Agentmail#197
Open
cooper-grc wants to merge 7 commits into
Open
Conversation
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
There was a problem hiding this comment.
2 issues found across 8 files
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="pkg/clients/agentmail.go">
<violation number="1" location="pkg/clients/agentmail.go:151">
P2: URL-encode pageToken before appending it to the query string; raw concatenation can break pagination when the token contains reserved characters.</violation>
</file>
<file name="pkg/sources/providers/agentmail.go">
<violation number="1" location="pkg/sources/providers/agentmail.go:122">
P2: Do not return sentinel size `0` for generated files in `ReadDir`; advertise real content size to prevent size-based readers from truncating or skipping content.
(Based on your team's feedback about avoiding fixed sentinel file sizes in directory metadata.) [FEEDBACK_USED]</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review, or fix all with cubic.
There was a problem hiding this comment.
4 issues found across 17 files (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="pkg/gateway/services/view_filters.go">
<violation number="1" location="pkg/gateway/services/view_filters.go:410">
P1: Combining `from` and `subject` into one space-joined AgentMail query breaks multi-field filtering because the provider performs literal substring matching on a single query string.</violation>
</file>
<file name="pkg/sources/queries/baml_client/baml_source_map.go">
<violation number="1" location="pkg/sources/queries/baml_client/baml_source_map.go:20">
P2: The embedded generated BAML map is out of sync with `baml_src`: it downgrades `generators.baml` to `0.218.1` while source is `0.220.0`. Keep generated artifacts consistent with source.</violation>
</file>
<file name="pkg/sources/providers/agentmail.go">
<violation number="1" location="pkg/sources/providers/agentmail.go:192">
P2: ExecuteQuery only reads the first page of inboxes, so results silently omit inboxes after the first 100.</violation>
<violation number="2" location="pkg/sources/providers/agentmail.go:206">
P1: Do not silently ignore inbox fetch errors in ExecuteQuery; returning partial/empty success can mask provider failures and miss events.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review, or fix all with cubic.
| if f.Subject != "" { | ||
| parts = append(parts, f.Subject) | ||
| } | ||
| spec := newSpec("agentmail", strings.Join(parts, " "), limit) |
There was a problem hiding this comment.
P1: Combining from and subject into one space-joined AgentMail query breaks multi-field filtering because the provider performs literal substring matching on a single query string.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At pkg/gateway/services/view_filters.go, line 410:
<comment>Combining `from` and `subject` into one space-joined AgentMail query breaks multi-field filtering because the provider performs literal substring matching on a single query string.</comment>
<file context>
@@ -387,6 +395,25 @@ func buildConfluenceFilter(raw json.RawMessage, limit int) (string, error) {
+ if f.Subject != "" {
+ parts = append(parts, f.Subject)
+ }
+ spec := newSpec("agentmail", strings.Join(parts, " "), limit)
+ if f.Inbox != "" {
+ spec["inbox_filter"] = f.Inbox
</file context>
| for _, id := range inboxIDs { | ||
| msgs, err := a.getCachedMessages(ctx, id) | ||
| if err != nil { | ||
| continue |
There was a problem hiding this comment.
P1: Do not silently ignore inbox fetch errors in ExecuteQuery; returning partial/empty success can mask provider failures and miss events.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At pkg/sources/providers/agentmail.go, line 206:
<comment>Do not silently ignore inbox fetch errors in ExecuteQuery; returning partial/empty success can mask provider failures and miss events.</comment>
<file context>
@@ -165,6 +167,165 @@ func (a *AgentMailProvider) Search(_ context.Context, _ *sources.ProviderContext
+ for _, id := range inboxIDs {
+ msgs, err := a.getCachedMessages(ctx, id)
+ if err != nil {
+ continue
+ }
+ all = append(all, msgs...)
</file context>
Suggested change
| continue | |
| return nil, err |
| "cron.baml": "class CronResult {\n cron_expr string @description(\"Standard 5-field cron expression: minute hour day_of_month month day_of_week\")\n}\n\nfunction ParseCronSchedule(description: string, timezone: string) -> CronResult {\n client SmallModelCerebras\n prompt #\"\n Convert the following human-readable schedule description into a standard\n 5-field cron expression (minute hour day_of_month month day_of_week).\n\n The user is in the {{ timezone }} timezone. All times in the description\n refer to {{ timezone }} local time. Output the cron expression in that\n local timezone (do NOT convert to UTC).\n\n Description: {{ description }}\n\n {{ ctx.output_format }}\n \"#\n}\n", | ||
| "generators.baml": "generator target {\n output_type \"go\"\n output_dir \"../\"\n version \"0.220.0\"\n\n // 'baml-cli generate' will run this after generating go code\n // This command will be run from within $output_dir/baml_client\n on_generate \"bash \\\"$(git rev-parse --show-toplevel)/bin/patch-baml-runtime.sh\\\" && gofmt -w . && goimports -w .\"\n\n // Go packages name as specified in go.mod\n // We need this to generate correct imports in the generated baml_client\n client_package_name \"github.com/beam-cloud/airstore/pkg/sources/queries\"\n}\n", | ||
| "smart_queries.baml": "// =============================================================================\n// Smart Query BAML Functions\n// Infer source-specific queries from folder/file names\n// =============================================================================\n\nclass GmailQueryResult {\n gmail_query string @description(\"Gmail search query using Gmail's operators\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {date}, {from}, {subject}. Example: {date}_{from}_{subject}_{id}.txt\")\n}\n\nclass GDriveQueryResult {\n gdrive_query string @description(\"Google Drive search query\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {name}, {date}. Example: {name}_{id}\")\n}\n\nclass NotionQueryResult {\n notion_query string @description(\"Notion search query text\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {title}, {date}. Example: {title}_{id}.md\")\n}\n\nclass GitHubQueryResult {\n github_query string @description(\"GitHub query - format depends on search_type\")\n search_type string @description(\"Type: repos, issues, prs, commits, releases, workflows, files, branches\")\n content_type string @description(\"Format: markdown (default), diff (PRs/commits), json (metadata), raw (files)\")\n limit int @description(\"Max results (default: 50)\")\n filename_format string @description(\"Filename template with placeholders like {number}, {title}, {sha}, {name}, {path}\")\n}\n\nclass SlackQueryResult {\n slack_query string @description(\"Slack search query using Slack's search syntax\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {channel}, {user}, {date}, {text}. Example: {date}_{channel}_{user}_{id}.txt\")\n}\n\nclass LinearQueryResult {\n linear_query string @description(\"Linear issue filter query using Linear's filter syntax\")\n search_type string @description(\"Type of search: 'issues' or 'projects'. Default: 'issues'\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {identifier}, {title}, {state}, {assignee}, {team}, {priority}, {date}. Example: {identifier}_{title}.md\")\n}\n\nclass PostHogQueryResult {\n posthog_query string @description(\"Search query string for PostHog API\")\n search_type string @description(\"Type of search: 'events', 'feature-flags', 'insights', 'cohorts'\")\n project_id int @description(\"PostHog project ID to search in. Use 0 to auto-select the first available project.\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Placeholders vary by search_type.\")\n}\n\nclass WebQueryResult {\n web_mode string @description(\"'map' to crawl a specific site, 'search' to find pages via web search. Default: map\")\n web_query string @description(\"For web_mode=map: the URL to crawl. For web_mode=search: the search query string.\")\n include_paths string[] @description(\"URL path patterns to include when web_mode=map (e.g. /cocktails/*). Empty array = all paths. Ignored for search.\")\n limit int @description(\"Max pages to discover (default: 100)\")\n filename_format string @description(\"Template for result filenames. Use {title}, {path}, {id}, {date}. Example: {title}_{id}.md\")\n}\n\n// -----------------------------------------------------------------------------\n// PostHog Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferPostHogQuery(name: string, guidance: string?) -> PostHogQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a PostHog query.\n\n SEARCH TYPES:\n 1. feature-flags — Feature flags (toggles). Use when name contains: flag, toggle, feature, rollout, experiment\n 2. insights — Saved queries, funnels, retention, trends. Use when name contains: insight, funnel, retention, trend, dashboard, chart, analytics\n 3. cohorts — User segments/groups. Use when name contains: cohort, segment, group, users, audience\n 4. events — Raw event data (default). Use when name contains: event, pageview, click, action, or nothing specific\n\n POSTHOG STANDARD EVENTS (start with $):\n $pageview, $pageleave, $autocapture, $screen, $exception, $rageclick, $web_vitals\n\n QUERY FORMAT:\n - For events: the event name to filter on (e.g., \"$pageview\", \"signup\", \"purchase\"). Empty string = all recent events.\n - For feature-flags/insights/cohorts: a search term to filter by name/key (e.g., \"beta\", \"onboarding\", \"retention\").\n\n ACTIVE FLAGS:\n If the name implies \"active\" or \"enabled\" flags (e.g., \"active-feature-flags\", \"enabled-flags\"), set search_type to \"feature-flags\" and use a broad query (empty or relevant keyword). The provider will filter for active=true.\n\n DEFAULTS:\n - project_id: 0 (auto-select first project)\n - limit: 50\n\n FILENAME FORMATS by search_type:\n - events: {date}_{event}_{id}.json\n - feature-flags: {key}_{id}.json\n - insights: {name}_{id}.json\n - cohorts: {name}_{id}.json\n\n Query examples:\n - \"active-feature-flags\" → search_type: feature-flags, query: \"\" (all flags, provider filters active)\n - \"pageview-events\" → search_type: events, query: \"$pageview\"\n - \"error-events\" → search_type: events, query: \"$exception\"\n - \"retention-insights\" → search_type: insights, query: \"retention\"\n - \"beta-users\" → search_type: cohorts, query: \"beta\"\n - \"signup-funnel\" → search_type: insights, query: \"signup funnel\"\n - \"recent-events\" → search_type: events, query: \"\"\n - \"experiment-flags\" → search_type: feature-flags, query: \"experiment\"\n - \"rage-clicks\" → search_type: events, query: \"$rageclick\"\n - \"web-vitals\" → search_type: events, query: \"$web_vitals\"\n - \"onboarding-cohort\" → search_type: cohorts, query: \"onboarding\"\n\n Folder/file name: {{ name }}\n {% if guidance %}\n\n Additional guidance from user: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Gmail Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGmailQuery(name: string, guidance: string?) -> GmailQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Gmail search query.\n \n IMPORTANT: Gmail search is FUZZY by default!\n - Bare words (no operator) search across subject, body, AND sender name/email\n - Use bare words for most searches - they match everywhere!\n - Gmail automatically handles plurals, common misspellings, and related terms\n \n CRITICAL: PREFER BARE WORDS OVER from: FOR FUZZY SENDER MATCHING!\n - from:company only matches if that word appears exactly in sender name/email\n - Bare words match sender, subject, AND body - much more flexible!\n - Many companies use subdomains like noreply@mail.company.com - bare words catch these better\n - ONLY use from: when you need STRICT sender matching (rare)\n \n CRITICAL SYNTAX RULES:\n - Multi-word operator values MUST be quoted: from:\"summit health\" NOT from:summit health\n - Bare multi-word phrases should also be quoted: \"lab results\" NOT lab results\n - Unquoted spaces mean AND: lab results = lab AND results (both must appear)\n - Use OR to match alternatives: mychart OR \"nyu langone\" OR \"summit health\"\n \n Gmail search operators (use sparingly - bare words are usually better!):\n - from:\"name or email\" - STRICT sender match (avoid for fuzzy matching!)\n - subject:\"phrase\" - matches subject line only\n - has:attachment, filename:pdf\n - is:unread, is:starred, is:important\n - newer_than:1d, newer_than:7d, older_than:30d\n \n QUERY STRATEGY:\n 1. For company/brand emails → USE BARE WORDS: nike OR adidas OR \"north face\"\n (This finds emails FROM them, ABOUT them, or MENTIONING them - much better coverage!)\n 2. For topics → bare words or quoted phrases: \"lab results\" OR prescription OR appointment\n 3. For people → from:firstname works but bare word also good: john OR from:john\n 4. For mixed (topic + senders) → all bare words: mychart OR nyu OR \"summit health\" OR prescription\n 5. For health/medical → medical OR health OR doctor OR appointment OR mychart OR prescription\n 6. When in doubt → ADD MORE ORs with related terms, synonyms, and variations!\n \n DEFAULTS:\n - limit: 50\n - filename_format: {date}_{from}_{subject}_{id}.txt\n \n Query examples:\n - \"invoices\" → invoice OR receipt OR bill OR statement\n - \"stripe-invoices\" → stripe invoice OR stripe receipt OR stripe payment\n - \"receipts\" → receipt OR invoice OR order OR confirmation OR purchase\n - \"meeting-notes\" → meeting OR agenda OR \"calendar invite\" OR standup OR sync\n - \"flight-confirmation\" → flight OR itinerary OR \"boarding pass\" OR airline OR booking\n - \"health-emails\" → medical OR health OR doctor OR appointment OR mychart OR prescription OR lab\n - \"from-john\" → from:john (explicit \"from\" in name = use the operator)\n - \"emails-from-acme-corp\" → from:\"acme corp\" (explicit \"from\" + multi-word = quoted operator)\n - \"unread\" → is:unread\n - \"project-alpha\" → \"project alpha\" (quoted phrase)\n - \"shipping\" → shipping OR tracking OR delivery OR fedex OR ups OR usps OR package\n - \"recent\" → newer_than:7d\n - \"amazon-orders\" → amazon order OR amazon shipment OR amazon delivery OR amazon purchase\n - \"uber-receipts\" → uber receipt OR uber ride OR lyft receipt OR lyft ride\n \n Guidance examples:\n - \"from nike\" → nike (bare word for fuzzy matching - catches nike.com, news@nike.com, etc.)\n - \"anything from summit health, nyu, mychart, health related\" \n → mychart OR nyu OR \"summit health\" OR medical OR appointment OR prescription OR lab\n (Use bare words with lots of ORs - they match sender names AND content!)\n - \"strictly only emails where the sender is john@company.com\" → from:john@company.com\n (Use from: ONLY when user explicitly wants strict sender matching)\n \n Filename format: {date}_{from}_{subject}_{id}.txt\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Gmail Query Evaluation and Refinement\n// -----------------------------------------------------------------------------\n\nclass GmailQueryEvaluation {\n is_satisfactory bool @description(\"True if results match user intent well enough\")\n refined_query string? @description(\"Improved Gmail query if not satisfactory, null if satisfactory\")\n reasoning string @description(\"Brief explanation of evaluation (1-2 sentences)\")\n}\n\nfunction EvaluateGmailQueryResults(\n original_guidance: string,\n generated_query: string,\n result_count: int,\n sample_results: string\n) -> GmailQueryEvaluation {\n client AnthropicClient\n prompt #\"\n You are a STRICT evaluator checking if Gmail search results actually match what the user asked for.\n Be critical - if results don't clearly match the user's intent, suggest a better query.\n \n USER'S REQUEST: {{ original_guidance }}\n \n CURRENT QUERY: {{ generated_query }}\n \n RESULTS: {{ result_count }} emails found\n \n SAMPLE RESULTS: {{ sample_results }}\n \n STRICT EVALUATION CHECKLIST:\n \n 1. ZERO RESULTS = AUTOMATIC FAIL\n - If result_count is 0, the query is broken. Mark as NOT satisfactory.\n - Common causes:\n * Unquoted multi-word values: from:summit health → from:\"summit health\"\n * Using from: when bare word would match better (from:bonobos might miss noreply@e.bonobos.com)\n \n 2. CHECK EACH THING THE USER MENTIONED:\n - If user mentioned specific senders/companies (e.g., \"bonobos\", \"amazon\", \"stripe\"):\n → Are emails FROM or ABOUT these in the results?\n → Note: bare words like \"bonobos\" are GOOD - they match sender, subject, AND body\n → If from: operator fails, suggest using bare word instead for fuzzy matching\n - If user mentioned topics (e.g., \"health\", \"medical\", \"appointments\"):\n → Do the subjects/snippets actually contain these words?\n → Generic results that don't mention the topics = NOT satisfactory.\n \n 3. LOOK FOR OBVIOUS MISMATCHES:\n - Results about completely unrelated topics = NOT satisfactory\n - Missing the main thing the user asked for = NOT satisfactory\n \n 4. QUERY IMPROVEMENTS TO SUGGEST:\n - from:bonobos (0 results) → bonobos (bare word matches subdomains like e.bonobos.com)\n - from:summit health → from:\"summit health\" (MUST quote multi-word)\n - Using from: when company uses weird email domains → use bare word instead\n - Not enough OR alternatives for fuzzy matching\n \n BE HARSH: If you're unsure whether results match, lean toward NOT satisfactory and suggest improvements.\n Only mark satisfactory if the results CLEARLY match what the user asked for.\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Google Drive Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGDriveQuery(name: string, guidance: string?) -> GDriveQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Google Drive files.list query (the `q` parameter).\n \n Output MUST be ONLY the query expression (no `q=`, no URL encoding).\n It MUST be valid Drive query syntax.\n \n Drive query syntax:\n - name contains 'text'\n - fullText contains 'text'\n - mimeType = 'application/pdf'\n - mimeType = 'application/vnd.google-apps.document'\n - mimeType = 'application/vnd.google-apps.spreadsheet'\n - mimeType = 'application/vnd.google-apps.presentation'\n - mimeType contains 'image'\n - sharedWithMe = true\n - 'me' in owners\n - 'email@example.com' in owners\n - 'email@example.com' in writers\n - 'email@example.com' in readers\n - starred = true\n - trashed = false\n - modifiedTime > '2024-01-01T00:00:00'\n - createdTime > '2024-01-01T00:00:00'\n - 'FOLDER_ID' in parents\n \n Combine with: and, or, not\n \n IMPORTANT:\n - Always include `trashed = false` unless the user explicitly asks for trash / deleted files.\n - Prefer `fullText contains` for topic/keyword folders (it searches name + indexed content).\n Use `name contains` when the intent is specifically filename-based.\n - Use RFC3339 timestamps for createdTime/modifiedTime. Prefer including a timezone, e.g. '2026-01-22T00:00:00Z'.\n \n DEFAULTS:\n - limit: 50\n - filename_format: {name}_{id}\n \n Query examples:\n - \"recent\" → trashed = false and modifiedTime > '2026-01-22T00:00:00Z'\n - \"shared\" → trashed = false and sharedWithMe = true\n - \"my-docs\" → trashed = false and 'me' in owners and mimeType = 'application/vnd.google-apps.document'\n - \"pdfs\" → trashed = false and mimeType = 'application/pdf'\n - \"documents\" → trashed = false and mimeType = 'application/vnd.google-apps.document'\n - \"spreadsheets\" → trashed = false and mimeType = 'application/vnd.google-apps.spreadsheet'\n - \"presentations\" → trashed = false and mimeType = 'application/vnd.google-apps.presentation'\n - \"images\" → trashed = false and mimeType contains 'image/'\n - \"starred\" → trashed = false and starred = true\n - \"college-docs\" → trashed = false and (fullText contains 'college' or name contains 'college') and mimeType = 'application/vnd.google-apps.document'\n - Guidance: \"created in the past week\" → use createdTime > '...'\n - Guidance: \"modified in the past week\" → use modifiedTime > '...'\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {name}, {date}, {created}, {mime_type}, {ext}\n - {name} is the basename (no extension)\n - {ext} is the extension (includes leading dot)\n - For documents: {name}_{id}\n - For dated files: {date}_{name}_{id}\n - For created-time focused queries: {created}_{name}_{id}\n - Default: {name}_{id}\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Notion Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferNotionQuery(name: string, guidance: string?) -> NotionQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Notion search query.\n \n Notion search is simple text-based. Extract key search terms from the folder name.\n Remove common words like \"my\", \"the\", \"all\", etc.\n \n DEFAULTS:\n - limit: 50\n - filename_format: {title}_{id}.md\n \n Query examples:\n - \"meeting-notes\" → meeting notes\n - \"project-docs\" → project docs\n - \"todo\" → todo\n - \"weekly-reports\" → weekly reports\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {title}, {type}, {date}, {created}\n - For notes: {title}_{id}.md\n - For dated pages: {date}_{title}_{id}.md\n - Default: {title}_{id}.md\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Slack Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferSlackQuery(name: string, guidance: string?) -> SlackQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Slack search query.\n \n Slack search supports various operators for finding messages across channels.\n The query should find relevant messages based on the folder name intent.\n \n SLACK SEARCH OPERATORS:\n - in:#channel-name - messages in a specific channel\n - in:@username - direct messages with a user\n - from:@username - messages from a specific user\n - from:me - messages you sent\n - to:me - messages sent to you (DMs)\n - has:link - messages containing links\n - has:reaction - messages with reactions\n - has:star - starred messages\n - has:pin - pinned messages\n - before:YYYY-MM-DD - messages before a date\n - after:YYYY-MM-DD - messages after a date\n - during:month - messages from a specific month (e.g., during:january)\n - on:YYYY-MM-DD - messages on a specific date\n - is:saved - saved messages\n \n QUERY STRATEGY:\n 1. For channel-specific queries → use in:#channel\n 2. For user-specific queries → use from:@user\n 3. For topic searches → use keywords directly\n 4. For time-based queries → use before:, after:, during:, on:\n 5. For content type → use has:link, has:reaction, etc.\n 6. Combine operators with spaces (implicit AND) or OR\n \n DEFAULTS:\n - limit: 50\n - filename_format: {date}_{channel}_{user}_{id}.txt\n \n Query examples:\n - \"engineering-updates\" → in:#engineering OR in:#engineering-updates\n - \"team-announcements\" → in:#announcements OR in:#general announcement\n - \"from-ceo\" → from:@ceo OR from:@founder\n - \"recent-links\" → has:link after:2026-01-01\n - \"pinned-messages\" → has:pin\n - \"design-feedback\" → in:#design feedback OR in:#design-reviews\n - \"standup-notes\" → standup OR \"daily standup\" OR in:#standup\n - \"customer-issues\" → in:#support OR in:#customer-success issue OR bug\n - \"product-discussions\" → in:#product OR product roadmap OR feature\n - \"onboarding\" → onboarding OR \"new hire\" OR welcome\n - \"decisions\" → decision OR decided OR \"we will\" OR approved\n - \"action-items\" → \"action item\" OR TODO OR \"follow up\" OR assigned\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {channel}, {user}, {date}, {text}\n - For channel messages: {date}_{channel}_{user}_{id}.txt\n - For user messages: {date}_{user}_{channel}_{id}.txt\n - Default: {date}_{channel}_{user}_{id}.txt\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Linear Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferLinearQuery(name: string, guidance: string?) -> LinearQueryResult {\n client AnthropicClient\n prompt #\"\n Convert natural language into a Linear search query.\n \n CORE PRINCIPLE: Text search is powerful - use it liberally. Only add filters when intent is unambiguous.\n \n SYNTAX:\n - Plain text: searches title + description (very flexible, use this a lot!)\n - assignee:name | assignee:me - who it's assigned to\n - no:assignee - unassigned issues\n - state:backlog | state:todo | state:\"in progress\" | state:done | state:canceled\n - priority:1 (urgent) | priority:2 (high) | priority:3 (medium) | priority:4 (low) \n - label:name - filter by label\n - no:label - unlabeled issues\n - team:KEY - filter by team key\n - project:name - filter by project name\n - creator:name - who created it\n - cycle:current | cycle:next | cycle:previous - filter by cycle\n - estimate:N - filter by point estimate\n - OR - combine alternatives (use generously!)\n \n STRATEGY - Think about what the user ACTUALLY wants:\n \n 1. TOPIC/FEATURE searches → mostly text search\n \"checkout bugs\" → checkout bug\n \"api performance\" → api performance\n \"mobile crashes\" → mobile crash\n \"auth issues\" → auth authentication login\n \n 2. PERSON searches → assignee filter + text backup\n \"john's tickets\" → assignee:john OR john\n \"sarah jones work\" → assignee:sarah OR assignee:jones OR sarah OR jones\n \"assigned to mike\" → assignee:mike\n \"unassigned issues\" → no:assignee\n \n 3. STATUS/PRIORITY searches → filters\n \"my open issues\" → assignee:me state:todo OR state:\"in progress\"\n \"urgent bugs\" → priority:1 bug\n \"blocked tickets\" → blocked OR blocking\n \"stuff in review\" → review\n \"in progress issues\" → state:\"in progress\"\n \n 4. PROJECT/CYCLE searches → project or cycle filter\n \"acme project\" → project:acme\n \"current sprint\" → cycle:current\n \"next sprint issues\" → cycle:next\n \"q1 roadmap\" → q1 roadmap\n \"enterprise features\" → enterprise\n \n 5. VAGUE/BROAD searches → text search with synonyms\n \"tech debt\" → \"tech debt\" OR refactor OR cleanup\n \"security\" → security vulnerability\n \"onboarding\" → onboarding OR \"getting started\" OR setup\n \n KEY RULES:\n - When in doubt, use text search - it's flexible\n - Use OR generously to catch variations \n - For multi-word names: assignee:first OR assignee:last (never the full name)\n - Add synonyms for common concepts\n - Don't over-filter - better to return more results than miss things\n \n SEARCH TYPE: \"issues\" (default) or \"projects\" (only if explicitly asking for projects list)\n \n DEFAULTS:\n - limit: 50\n - filename_format: {identifier}_{title}.md\n \n CRITICAL: filename_format MUST contain placeholders like {identifier}, {title}, etc.\n NEVER return a literal filename like \"test.md\" or the folder name as the format.\n The format is a TEMPLATE applied to every result, so it MUST use placeholders.\n \n Folder/file name: {{ name }}\n {% if guidance %}\n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// GitHub Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGitHubQuery(name: string, guidance: string?) -> GitHubQueryResult {\n client AnthropicClient\n prompt #\"\n Convert natural language into a GitHub query. Think about what an agent would need.\n \n ═══════════════════════════════════════════════════════════════════════════════\n SEARCH TYPES & QUERY FORMATS\n ═══════════════════════════════════════════════════════════════════════════════\n \n 1. REPOS (search_type: \"repos\")\n Listing: list:org/ORGNAME or list:user/USERNAME [type:private|public]\n Search: language:go stars:>100 topic:ml org:company\n \n Examples:\n - \"beam-cloud repos\" → list:org/beam-cloud\n - \"private repos in acme\" → list:org/acme type:private\n - \"popular go projects\" → language:go stars:>1000\n \n 2. ISSUES (search_type: \"issues\") \n Query: repo:owner/name [is:open|closed] [label:X] [assignee:X] [author:X]\n \n Examples:\n - \"open bugs in beta9\" → repo:beam-cloud/beta9 is:open label:bug\n - \"my issues\" → assignee:@me is:open\n - \"security issues in react\" → repo:facebook/react label:security\n \n 3. PULL REQUESTS (search_type: \"prs\")\n Query: repo:owner/name [is:open|closed|merged] [author:X] [review:approved]\n \n Examples:\n - \"open PRs in beta9\" → repo:beam-cloud/beta9 is:open\n - \"merged PRs this week\" → repo:owner/repo is:merged merged:>2026-01-26\n - \"PRs needing review\" → repo:owner/repo is:open review:none\n \n 4. COMMITS (search_type: \"commits\")\n Query: repo:owner/name [branch:X] [author:X] [since:DATE] [path:X]\n \n Examples:\n - \"recent commits in beta9\" → repo:beam-cloud/beta9\n - \"commits to main this week\" → repo:owner/repo branch:main since:2026-01-26\n - \"john's commits\" → repo:owner/repo author:john\n - \"changes to src/api\" → repo:owner/repo path:src/api\n \n 5. RELEASES (search_type: \"releases\")\n Query: repo:owner/name [latest] [prerelease:true|false]\n \n Examples:\n - \"beta9 releases\" → repo:beam-cloud/beta9\n - \"latest release of kubernetes\" → repo:kubernetes/kubernetes latest\n - \"stable releases only\" → repo:owner/repo prerelease:false\n \n 6. WORKFLOW RUNS (search_type: \"workflows\")\n Query: repo:owner/name [status:success|failure|in_progress] [branch:X] [event:push|pr]\n \n Examples:\n - \"failed CI in beta9\" → repo:beam-cloud/beta9 status:failure\n - \"recent builds\" → repo:owner/repo\n - \"PR checks\" → repo:owner/repo event:pull_request\n \n 7. FILES (search_type: \"files\")\n Query: repo:owner/name path:PATH [ref:branch|tag|sha]\n \n Examples:\n - \"beta9 readme\" → repo:beam-cloud/beta9 path:README.md\n - \"package.json in react\" → repo:facebook/react path:package.json\n - \"config files\" → repo:owner/repo path:*.yaml ref:main\n - \"src directory\" → repo:owner/repo path:src/\n \n 8. BRANCHES (search_type: \"branches\")\n Query: repo:owner/name [protected:true|false]\n \n Examples:\n - \"beta9 branches\" → repo:beam-cloud/beta9\n - \"protected branches\" → repo:owner/repo protected:true\n \n ═══════════════════════════════════════════════════════════════════════════════\n CONTENT TYPES (how to format the content)\n ═══════════════════════════════════════════════════════════════════════════════\n \n - \"markdown\" (default) - Rich formatted content, best for agents to read\n - \"diff\" - Code changes (for PRs/commits)\n - \"json\" - Structured metadata\n - \"raw\" - Unprocessed file content (for files)\n \n ═══════════════════════════════════════════════════════════════════════════════\n FILENAME FORMATS (use appropriate placeholders)\n ═══════════════════════════════════════════════════════════════════════════════\n \n Repos: {name}.md or {full_name}.json\n Issues: {number}_{title}.md\n PRs: {number}_{title}.md or {number}_{title}_diff.md\n Commits: {sha_short}_{message}.md\n Releases: {tag}_{name}.md\n Workflows: {id}_{name}_{status}.md\n Files: {path} (preserve original path/name)\n Branches: {name}.md\n \n ═══════════════════════════════════════════════════════════════════════════════\n SMART DETECTION\n ═══════════════════════════════════════════════════════════════════════════════\n \n Detect intent from keywords:\n - \"repos/repositories/projects\" in org/user → repos listing\n - \"prs/pulls/pull requests/merge requests\" → prs\n - \"issues/bugs/tickets/tasks\" → issues\n - \"commits/history/changes/log\" → commits \n - \"releases/versions/tags\" → releases\n - \"ci/cd/builds/workflows/actions/checks\" → workflows\n - \"readme/config/file/code\" → files\n - \"branches\" → branches\n - \"diff/patch/changes\" → content_type: diff\n \n Folder/file name: {{ name }}\n {% if guidance %}\n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Tests\n// -----------------------------------------------------------------------------\n\ntest TestGitHubOpenPRs {\n functions [InferGitHubQuery]\n args {\n name \"open-prs-beta9\"\n guidance null\n }\n}\n\ntest TestGitHubIssues {\n functions [InferGitHubQuery]\n args {\n name \"beta9-bugs\"\n guidance \"show open bugs in the beta9 repo\"\n }\n}\n\ntest TestGitHubRepos {\n functions [InferGitHubQuery]\n args {\n name \"go-cli-tools\"\n guidance null\n }\n}\n\ntest TestGmailUnread {\n functions [InferGmailQuery]\n args {\n name \"unread-emails\"\n guidance null\n }\n}\n\ntest TestGmailFromPerson {\n functions [InferGmailQuery]\n args {\n name \"from-eli\"\n guidance null\n }\n}\n\ntest TestGmailWithGuidance {\n functions [InferGmailQuery]\n args {\n name \"important-emails\"\n guidance \"Only from the last 7 days, max 100 results\"\n }\n}\n\ntest TestGmailInvoices {\n functions [InferGmailQuery]\n args {\n name \"invoices\"\n guidance null\n }\n}\n\ntest TestGmailMeetingNotes {\n functions [InferGmailQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestGmailFlightConfirmation {\n functions [InferGmailQuery]\n args {\n name \"flight-confirmations\"\n guidance null\n }\n}\n\ntest TestGmailReceipts {\n functions [InferGmailQuery]\n args {\n name \"receipts\"\n guidance null\n }\n}\n\ntest TestGmailProjectAlpha {\n functions [InferGmailQuery]\n args {\n name \"project-alpha-emails\"\n guidance null\n }\n}\n\ntest TestGmailShippingUpdates {\n functions [InferGmailQuery]\n args {\n name \"shipping-updates\"\n guidance null\n }\n}\n\ntest TestGmailHealthWithGuidance {\n functions [InferGmailQuery]\n args {\n name \"Medical Documents\"\n guidance \"anything from: summit health, nyu, anything health related basically, mychart\"\n }\n}\n\ntest TestGmailMultiWordSender {\n functions [InferGmailQuery]\n args {\n name \"acme-corp-emails\"\n guidance \"from acme corp and related companies\"\n }\n}\n\ntest TestGmailFuzzySenderNike {\n functions [InferGmailQuery]\n args {\n name \"nike-emails\"\n guidance \"from nike\"\n }\n}\n\ntest TestGmailFuzzySenderAmazon {\n functions [InferGmailQuery]\n args {\n name \"amazon-orders\"\n guidance null\n }\n}\n\ntest TestGDriveShared {\n functions [InferGDriveQuery]\n args {\n name \"shared-with-me\"\n guidance null\n }\n}\n\ntest TestGDrivePdfs {\n functions [InferGDriveQuery]\n args {\n name \"pdfs\"\n guidance null\n }\n}\n\ntest TestGDriveCollegeDocs {\n functions [InferGDriveQuery]\n args {\n name \"college docs\"\n guidance null\n }\n}\n\ntest TestGDriveRecentDocsWithGuidance {\n functions [InferGDriveQuery]\n args {\n name \"recent-docs\"\n guidance \"any docs created in the past week\"\n }\n}\n\ntest TestGDriveOwnedByMe {\n functions [InferGDriveQuery]\n args {\n name \"my docs\"\n guidance null\n }\n}\n\ntest TestNotionMeetings {\n functions [InferNotionQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestEvaluateGmailResults_NoResults {\n functions [EvaluateGmailQueryResults]\n args {\n original_guidance \"anything from summit health, nyu, mychart, health related\"\n generated_query \"from:summit health OR from:nyu OR from:mychart\"\n result_count 0\n sample_results \"[]\"\n }\n}\n\ntest TestEvaluateGmailResults_GoodResults {\n functions [EvaluateGmailQueryResults]\n args {\n original_guidance \"invoices from stripe\"\n generated_query \"invoice from:stripe\"\n result_count 15\n sample_results \"[{\\\"from\\\":\\\"Stripe\\\",\\\"subject\\\":\\\"Your invoice from Acme Corp\\\",\\\"snippet\\\":\\\"Invoice #12345...\\\"},{\\\"from\\\":\\\"Stripe Billing\\\",\\\"subject\\\":\\\"Invoice for January\\\",\\\"snippet\\\":\\\"Amount due: $99\\\"}]\"\n }\n}\n\ntest TestSlackEngineering {\n functions [InferSlackQuery]\n args {\n name \"engineering-updates\"\n guidance null\n }\n}\n\ntest TestSlackFromCEO {\n functions [InferSlackQuery]\n args {\n name \"from-ceo\"\n guidance \"messages from our CEO\"\n }\n}\n\ntest TestSlackRecentLinks {\n functions [InferSlackQuery]\n args {\n name \"recent-links\"\n guidance null\n }\n}\n\ntest TestSlackStandups {\n functions [InferSlackQuery]\n args {\n name \"standup-notes\"\n guidance null\n }\n}\n\ntest TestSlackDecisions {\n functions [InferSlackQuery]\n args {\n name \"team-decisions\"\n guidance \"important decisions made by the team\"\n }\n}\n\ntest TestLinearMyIssues {\n functions [InferLinearQuery]\n args {\n name \"my-issues\"\n guidance null\n }\n}\n\ntest TestLinearMyBugs {\n functions [InferLinearQuery]\n args {\n name \"my-bugs\"\n guidance \"bugs assigned to me\"\n }\n}\n\ntest TestLinearUrgent {\n functions [InferLinearQuery]\n args {\n name \"urgent-issues\"\n guidance null\n }\n}\n\ntest TestLinearTeamBacklog {\n functions [InferLinearQuery]\n args {\n name \"eng-backlog\"\n guidance \"engineering team backlog\"\n }\n}\n\ntest TestLinearInProgress {\n functions [InferLinearQuery]\n args {\n name \"in-progress\"\n guidance null\n }\n}\n\ntest TestPostHogActiveFlags {\n functions [InferPostHogQuery]\n args {\n name \"active-feature-flags\"\n guidance null\n }\n}\n\ntest TestPostHogPageviewEvents {\n functions [InferPostHogQuery]\n args {\n name \"pageview-events\"\n guidance null\n }\n}\n\ntest TestPostHogRetentionInsights {\n functions [InferPostHogQuery]\n args {\n name \"retention-insights\"\n guidance null\n }\n}\n\ntest TestPostHogBetaUsers {\n functions [InferPostHogQuery]\n args {\n name \"beta-users\"\n guidance null\n }\n}\n\ntest TestPostHogErrorEvents {\n functions [InferPostHogQuery]\n args {\n name \"error-events\"\n guidance \"exception and error events from the past week\"\n }\n}\n\ntest TestPostHogSignupFunnel {\n functions [InferPostHogQuery]\n args {\n name \"signup-funnel\"\n guidance null\n }\n}\n\n// -----------------------------------------------------------------------------\n// Confluence Query Inference\n// -----------------------------------------------------------------------------\n\nclass ConfluenceQueryResult {\n cql_query string @description(\"Confluence Query Language (CQL) query string\")\n content_type string @description(\"Content type filter: 'page', 'blogpost', or 'all'. Default: 'all'\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {title}, {space}, {date}, {type}. Example: {space}_{title}_{id}.md\")\n}\n\nfunction InferConfluenceQuery(name: string, guidance: string?) -> ConfluenceQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Confluence CQL (Confluence Query Language) query.\n\n CQL SYNTAX:\n - type=page — pages only\n - type=blogpost — blog posts only\n - space=KEY — filter by space key (e.g., space=DEV, space=ENG)\n - title~\"text\" — title contains text (fuzzy match)\n - text~\"text\" — full-text search (searches title + body)\n - label=\"name\" — filter by label\n - creator=\"username\" — filter by creator\n - contributor=\"username\" — filter by contributor\n - lastModified>now(\"-7d\") — modified in the last 7 days\n - lastModified>now(\"-30d\") — modified in the last 30 days\n - lastModified>\"2025-01-01\" — modified after a specific date\n - created>now(\"-7d\") — created in the last 7 days\n - ancestor=12345 — pages under a specific parent\n - AND, OR — combine conditions\n - ORDER BY lastModified DESC, title ASC — sorting\n\n QUERY STRATEGY:\n 1. For topic searches → use text~ for full-text search\n 2. For space-specific → use space=KEY\n 3. For recent content → use lastModified>now(\"-Nd\")\n 4. For labeled content → use label=\"name\"\n 5. For specific content types → use type=page or type=blogpost\n 6. Combine operators with AND/OR\n\n DEFAULTS:\n - content_type: \"all\"\n - limit: 50\n - filename_format: {space}_{title}_{id}.md\n\n Query examples:\n - \"recent-pages\" → lastModified>now(\"-7d\") ORDER BY lastModified DESC, content_type: \"page\"\n - \"engineering-docs\" → (space=ENG OR text~\"engineering\") AND type=page ORDER BY lastModified DESC\n - \"kubernetes-pages\" → text~\"kubernetes\" AND type=page ORDER BY lastModified DESC\n - \"dev-blog-posts\" → space=DEV AND type=blogpost ORDER BY lastModified DESC\n - \"important-docs\" → label=\"important\" ORDER BY lastModified DESC\n - \"api-documentation\" → text~\"api\" AND (label=\"documentation\" OR title~\"api\") ORDER BY lastModified DESC\n - \"onboarding\" → text~\"onboarding\" OR title~\"onboarding\" ORDER BY lastModified DESC\n - \"architecture-decisions\" → text~\"architecture\" OR label=\"adr\" ORDER BY lastModified DESC\n - \"meeting-notes\" → title~\"meeting\" OR label=\"meeting-notes\" ORDER BY lastModified DESC\n - \"recent-blog-posts\" → type=blogpost AND lastModified>now(\"-30d\") ORDER BY lastModified DESC\n - \"project-alpha\" → text~\"project alpha\" ORDER BY lastModified DESC\n\n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {title}, {space}, {date}, {type}\n - For space-specific: {title}_{id}.md\n - For cross-space: {space}_{title}_{id}.md\n - For dated content: {date}_{title}_{id}.md\n - Default: {space}_{title}_{id}.md\n\n Folder/file name: {{ name }}\n {% if guidance %}\n\n Additional guidance from user: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Web Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferWebQuery(name: string, guidance: string?) -> WebQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder name into a web query. Two modes:\n\n web_mode=map — crawl a specific website. web_query must be a valid https:// URL.\n web_mode=search — web search. web_query is a search query string (not a URL).\n\n Use web_mode=map when guidance contains a URL or the name implies a specific site.\n Use web_mode=search when the name describes a topic, question, or information need.\n\n EXAMPLES:\n - \"hennessy-cocktails\" (guidance: \"https://www.hennessy.com/en-us/cocktails\")\n → web_mode: \"map\", web_query: \"https://www.hennessy.com/en-us/cocktails\", include_paths: [\"/en-us/cocktails/*\"], limit: 100\n - \"react-docs\"\n → web_mode: \"map\", web_query: \"https://react.dev/reference\", include_paths: [\"/reference/*\"], limit: 100\n - \"python-tutorial\"\n → web_mode: \"map\", web_query: \"https://docs.python.org/3/tutorial/\", include_paths: [\"/3/tutorial/*\"], limit: 50\n - \"latest AI news\"\n → web_mode: \"search\", web_query: \"latest AI news 2025\", include_paths: [], limit: 20\n - \"best rust web frameworks\"\n → web_mode: \"search\", web_query: \"best rust web frameworks comparison\", include_paths: [], limit: 15\n - \"climate change research papers\"\n → web_mode: \"search\", web_query: \"climate change research papers 2025\", include_paths: [], limit: 20\n\n Defaults: web_mode=map, limit=100, filename_format={title}_{id}.md\n\n Folder name: {{ name }}\n {% if guidance %}\n User guidance: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Web Query Tests\n// -----------------------------------------------------------------------------\n\ntest TestWebQueryWithGuidance {\n functions [InferWebQuery]\n args {\n name \"hennessy-cocktails\"\n guidance \"https://www.hennessy.com/en-us/cocktails?page=1\"\n }\n}\n\ntest TestWebQueryDocs {\n functions [InferWebQuery]\n args {\n name \"react-docs\"\n guidance null\n }\n}\n\ntest TestWebQueryBlog {\n functions [InferWebQuery]\n args {\n name \"openai-blog\"\n guidance null\n }\n}\n\ntest TestWebQuerySearch {\n functions [InferWebQuery]\n args {\n name \"latest AI news\"\n guidance null\n }\n}\n\ntest TestWebQuerySearchTopic {\n functions [InferWebQuery]\n args {\n name \"best rust web frameworks\"\n guidance null\n }\n}\n\n// -----------------------------------------------------------------------------\n// Confluence Query Tests\n// -----------------------------------------------------------------------------\n\ntest TestConfluenceRecentPages {\n functions [InferConfluenceQuery]\n args {\n name \"recent-pages\"\n guidance null\n }\n}\n\ntest TestConfluenceEngDocs {\n functions [InferConfluenceQuery]\n args {\n name \"engineering-docs\"\n guidance \"documentation from the engineering space\"\n }\n}\n\ntest TestConfluenceKubernetes {\n functions [InferConfluenceQuery]\n args {\n name \"kubernetes-pages\"\n guidance null\n }\n}\n\ntest TestConfluenceMeetingNotes {\n functions [InferConfluenceQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestConfluenceArchitectureDecisions {\n functions [InferConfluenceQuery]\n args {\n name \"architecture-decisions\"\n guidance \"ADRs and architecture decision records\"\n }\n}\n", | ||
| "generators.baml": "generator target {\n output_type \"go\"\n output_dir \"../\"\n version \"0.218.1\"\n\n // 'baml-cli generate' will run this after generating go code\n // This command will be run from within $output_dir/baml_client\n on_generate \"bash \\\"$(git rev-parse --show-toplevel)/bin/patch-baml-runtime.sh\\\" && gofmt -w . && goimports -w .\"\n\n // Go packages name as specified in go.mod\n // We need this to generate correct imports in the generated baml_client\n client_package_name \"github.com/beam-cloud/airstore/pkg/sources/queries\"\n}\n", |
There was a problem hiding this comment.
P2: The embedded generated BAML map is out of sync with baml_src: it downgrades generators.baml to 0.218.1 while source is 0.220.0. Keep generated artifacts consistent with source.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At pkg/sources/queries/baml_client/baml_source_map.go, line 20:
<comment>The embedded generated BAML map is out of sync with `baml_src`: it downgrades `generators.baml` to `0.218.1` while source is `0.220.0`. Keep generated artifacts consistent with source.</comment>
<file context>
@@ -17,8 +17,8 @@ var file_map = map[string]string{
"cron.baml": "class CronResult {\n cron_expr string @description(\"Standard 5-field cron expression: minute hour day_of_month month day_of_week\")\n}\n\nfunction ParseCronSchedule(description: string, timezone: string) -> CronResult {\n client SmallModelCerebras\n prompt #\"\n Convert the following human-readable schedule description into a standard\n 5-field cron expression (minute hour day_of_month month day_of_week).\n\n The user is in the {{ timezone }} timezone. All times in the description\n refer to {{ timezone }} local time. Output the cron expression in that\n local timezone (do NOT convert to UTC).\n\n Description: {{ description }}\n\n {{ ctx.output_format }}\n \"#\n}\n",
- "generators.baml": "generator target {\n output_type \"go\"\n output_dir \"../\"\n version \"0.220.0\"\n\n // 'baml-cli generate' will run this after generating go code\n // This command will be run from within $output_dir/baml_client\n on_generate \"bash \\\"$(git rev-parse --show-toplevel)/bin/patch-baml-runtime.sh\\\" && gofmt -w . && goimports -w .\"\n\n // Go packages name as specified in go.mod\n // We need this to generate correct imports in the generated baml_client\n client_package_name \"github.com/beam-cloud/airstore/pkg/sources/queries\"\n}\n",
- "smart_queries.baml": "// =============================================================================\n// Smart Query BAML Functions\n// Infer source-specific queries from folder/file names\n// =============================================================================\n\nclass GmailQueryResult {\n gmail_query string @description(\"Gmail search query using Gmail's operators\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {date}, {from}, {subject}. Example: {date}_{from}_{subject}_{id}.txt\")\n}\n\nclass GDriveQueryResult {\n gdrive_query string @description(\"Google Drive search query\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {name}, {date}. Example: {name}_{id}\")\n}\n\nclass NotionQueryResult {\n notion_query string @description(\"Notion search query text\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {title}, {date}. Example: {title}_{id}.md\")\n}\n\nclass GitHubQueryResult {\n github_query string @description(\"GitHub query - format depends on search_type\")\n search_type string @description(\"Type: repos, issues, prs, commits, releases, workflows, files, branches\")\n content_type string @description(\"Format: markdown (default), diff (PRs/commits), json (metadata), raw (files)\")\n limit int @description(\"Max results (default: 50)\")\n filename_format string @description(\"Filename template with placeholders like {number}, {title}, {sha}, {name}, {path}\")\n}\n\nclass SlackQueryResult {\n slack_query string @description(\"Slack search query using Slack's search syntax\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {channel}, {user}, {date}, {text}. Example: {date}_{channel}_{user}_{id}.txt\")\n}\n\nclass LinearQueryResult {\n linear_query string @description(\"Linear issue filter query using Linear's filter syntax\")\n search_type string @description(\"Type of search: 'issues' or 'projects'. Default: 'issues'\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {identifier}, {title}, {state}, {assignee}, {team}, {priority}, {date}. Example: {identifier}_{title}.md\")\n}\n\nclass PostHogQueryResult {\n posthog_query string @description(\"Search query string for PostHog API\")\n search_type string @description(\"Type of search: 'events', 'feature-flags', 'insights', 'cohorts'\")\n project_id int @description(\"PostHog project ID to search in. Use 0 to auto-select the first available project.\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Placeholders vary by search_type.\")\n}\n\nclass WebQueryResult {\n web_mode string @description(\"'map' to crawl a specific site, 'search' to find pages via web search. Default: map\")\n web_query string @description(\"For web_mode=map: the URL to crawl. For web_mode=search: the search query string.\")\n include_paths string[] @description(\"URL path patterns to include when web_mode=map (e.g. /cocktails/*). Empty array = all paths. Ignored for search.\")\n limit int @description(\"Max pages to discover (default: 100)\")\n filename_format string @description(\"Template for result filenames. Use {title}, {path}, {id}, {date}. Example: {title}_{id}.md\")\n}\n\n// -----------------------------------------------------------------------------\n// PostHog Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferPostHogQuery(name: string, guidance: string?) -> PostHogQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a PostHog query.\n\n SEARCH TYPES:\n 1. feature-flags — Feature flags (toggles). Use when name contains: flag, toggle, feature, rollout, experiment\n 2. insights — Saved queries, funnels, retention, trends. Use when name contains: insight, funnel, retention, trend, dashboard, chart, analytics\n 3. cohorts — User segments/groups. Use when name contains: cohort, segment, group, users, audience\n 4. events — Raw event data (default). Use when name contains: event, pageview, click, action, or nothing specific\n\n POSTHOG STANDARD EVENTS (start with $):\n $pageview, $pageleave, $autocapture, $screen, $exception, $rageclick, $web_vitals\n\n QUERY FORMAT:\n - For events: the event name to filter on (e.g., \"$pageview\", \"signup\", \"purchase\"). Empty string = all recent events.\n - For feature-flags/insights/cohorts: a search term to filter by name/key (e.g., \"beta\", \"onboarding\", \"retention\").\n\n ACTIVE FLAGS:\n If the name implies \"active\" or \"enabled\" flags (e.g., \"active-feature-flags\", \"enabled-flags\"), set search_type to \"feature-flags\" and use a broad query (empty or relevant keyword). The provider will filter for active=true.\n\n DEFAULTS:\n - project_id: 0 (auto-select first project)\n - limit: 50\n\n FILENAME FORMATS by search_type:\n - events: {date}_{event}_{id}.json\n - feature-flags: {key}_{id}.json\n - insights: {name}_{id}.json\n - cohorts: {name}_{id}.json\n\n Query examples:\n - \"active-feature-flags\" → search_type: feature-flags, query: \"\" (all flags, provider filters active)\n - \"pageview-events\" → search_type: events, query: \"$pageview\"\n - \"error-events\" → search_type: events, query: \"$exception\"\n - \"retention-insights\" → search_type: insights, query: \"retention\"\n - \"beta-users\" → search_type: cohorts, query: \"beta\"\n - \"signup-funnel\" → search_type: insights, query: \"signup funnel\"\n - \"recent-events\" → search_type: events, query: \"\"\n - \"experiment-flags\" → search_type: feature-flags, query: \"experiment\"\n - \"rage-clicks\" → search_type: events, query: \"$rageclick\"\n - \"web-vitals\" → search_type: events, query: \"$web_vitals\"\n - \"onboarding-cohort\" → search_type: cohorts, query: \"onboarding\"\n\n Folder/file name: {{ name }}\n {% if guidance %}\n\n Additional guidance from user: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Gmail Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGmailQuery(name: string, guidance: string?) -> GmailQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Gmail search query.\n \n IMPORTANT: Gmail search is FUZZY by default!\n - Bare words (no operator) search across subject, body, AND sender name/email\n - Use bare words for most searches - they match everywhere!\n - Gmail automatically handles plurals, common misspellings, and related terms\n \n CRITICAL: PREFER BARE WORDS OVER from: FOR FUZZY SENDER MATCHING!\n - from:company only matches if that word appears exactly in sender name/email\n - Bare words match sender, subject, AND body - much more flexible!\n - Many companies use subdomains like noreply@mail.company.com - bare words catch these better\n - ONLY use from: when you need STRICT sender matching (rare)\n \n CRITICAL SYNTAX RULES:\n - Multi-word operator values MUST be quoted: from:\"summit health\" NOT from:summit health\n - Bare multi-word phrases should also be quoted: \"lab results\" NOT lab results\n - Unquoted spaces mean AND: lab results = lab AND results (both must appear)\n - Use OR to match alternatives: mychart OR \"nyu langone\" OR \"summit health\"\n \n Gmail search operators (use sparingly - bare words are usually better!):\n - from:\"name or email\" - STRICT sender match (avoid for fuzzy matching!)\n - subject:\"phrase\" - matches subject line only\n - has:attachment, filename:pdf\n - is:unread, is:starred, is:important\n - newer_than:1d, newer_than:7d, older_than:30d\n \n QUERY STRATEGY:\n 1. For company/brand emails → USE BARE WORDS: nike OR adidas OR \"north face\"\n (This finds emails FROM them, ABOUT them, or MENTIONING them - much better coverage!)\n 2. For topics → bare words or quoted phrases: \"lab results\" OR prescription OR appointment\n 3. For people → from:firstname works but bare word also good: john OR from:john\n 4. For mixed (topic + senders) → all bare words: mychart OR nyu OR \"summit health\" OR prescription\n 5. For health/medical → medical OR health OR doctor OR appointment OR mychart OR prescription\n 6. When in doubt → ADD MORE ORs with related terms, synonyms, and variations!\n \n DEFAULTS:\n - limit: 50\n - filename_format: {date}_{from}_{subject}_{id}.txt\n \n Query examples:\n - \"invoices\" → invoice OR receipt OR bill OR statement\n - \"stripe-invoices\" → stripe invoice OR stripe receipt OR stripe payment\n - \"receipts\" → receipt OR invoice OR order OR confirmation OR purchase\n - \"meeting-notes\" → meeting OR agenda OR \"calendar invite\" OR standup OR sync\n - \"flight-confirmation\" → flight OR itinerary OR \"boarding pass\" OR airline OR booking\n - \"health-emails\" → medical OR health OR doctor OR appointment OR mychart OR prescription OR lab\n - \"from-john\" → from:john (explicit \"from\" in name = use the operator)\n - \"emails-from-acme-corp\" → from:\"acme corp\" (explicit \"from\" + multi-word = quoted operator)\n - \"unread\" → is:unread\n - \"project-alpha\" → \"project alpha\" (quoted phrase)\n - \"shipping\" → shipping OR tracking OR delivery OR fedex OR ups OR usps OR package\n - \"recent\" → newer_than:7d\n - \"amazon-orders\" → amazon order OR amazon shipment OR amazon delivery OR amazon purchase\n - \"uber-receipts\" → uber receipt OR uber ride OR lyft receipt OR lyft ride\n \n Guidance examples:\n - \"from nike\" → nike (bare word for fuzzy matching - catches nike.com, news@nike.com, etc.)\n - \"anything from summit health, nyu, mychart, health related\" \n → mychart OR nyu OR \"summit health\" OR medical OR appointment OR prescription OR lab\n (Use bare words with lots of ORs - they match sender names AND content!)\n - \"strictly only emails where the sender is john@company.com\" → from:john@company.com\n (Use from: ONLY when user explicitly wants strict sender matching)\n \n Filename format: {date}_{from}_{subject}_{id}.txt\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Gmail Query Evaluation and Refinement\n// -----------------------------------------------------------------------------\n\nclass GmailQueryEvaluation {\n is_satisfactory bool @description(\"True if results match user intent well enough\")\n refined_query string? @description(\"Improved Gmail query if not satisfactory, null if satisfactory\")\n reasoning string @description(\"Brief explanation of evaluation (1-2 sentences)\")\n}\n\nfunction EvaluateGmailQueryResults(\n original_guidance: string,\n generated_query: string,\n result_count: int,\n sample_results: string\n) -> GmailQueryEvaluation {\n client AnthropicClient\n prompt #\"\n You are a STRICT evaluator checking if Gmail search results actually match what the user asked for.\n Be critical - if results don't clearly match the user's intent, suggest a better query.\n \n USER'S REQUEST: {{ original_guidance }}\n \n CURRENT QUERY: {{ generated_query }}\n \n RESULTS: {{ result_count }} emails found\n \n SAMPLE RESULTS: {{ sample_results }}\n \n STRICT EVALUATION CHECKLIST:\n \n 1. ZERO RESULTS = AUTOMATIC FAIL\n - If result_count is 0, the query is broken. Mark as NOT satisfactory.\n - Common causes:\n * Unquoted multi-word values: from:summit health → from:\"summit health\"\n * Using from: when bare word would match better (from:bonobos might miss noreply@e.bonobos.com)\n \n 2. CHECK EACH THING THE USER MENTIONED:\n - If user mentioned specific senders/companies (e.g., \"bonobos\", \"amazon\", \"stripe\"):\n → Are emails FROM or ABOUT these in the results?\n → Note: bare words like \"bonobos\" are GOOD - they match sender, subject, AND body\n → If from: operator fails, suggest using bare word instead for fuzzy matching\n - If user mentioned topics (e.g., \"health\", \"medical\", \"appointments\"):\n → Do the subjects/snippets actually contain these words?\n → Generic results that don't mention the topics = NOT satisfactory.\n \n 3. LOOK FOR OBVIOUS MISMATCHES:\n - Results about completely unrelated topics = NOT satisfactory\n - Missing the main thing the user asked for = NOT satisfactory\n \n 4. QUERY IMPROVEMENTS TO SUGGEST:\n - from:bonobos (0 results) → bonobos (bare word matches subdomains like e.bonobos.com)\n - from:summit health → from:\"summit health\" (MUST quote multi-word)\n - Using from: when company uses weird email domains → use bare word instead\n - Not enough OR alternatives for fuzzy matching\n \n BE HARSH: If you're unsure whether results match, lean toward NOT satisfactory and suggest improvements.\n Only mark satisfactory if the results CLEARLY match what the user asked for.\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Google Drive Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGDriveQuery(name: string, guidance: string?) -> GDriveQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Google Drive files.list query (the `q` parameter).\n \n Output MUST be ONLY the query expression (no `q=`, no URL encoding).\n It MUST be valid Drive query syntax.\n \n Drive query syntax:\n - name contains 'text'\n - fullText contains 'text'\n - mimeType = 'application/pdf'\n - mimeType = 'application/vnd.google-apps.document'\n - mimeType = 'application/vnd.google-apps.spreadsheet'\n - mimeType = 'application/vnd.google-apps.presentation'\n - mimeType contains 'image'\n - sharedWithMe = true\n - 'me' in owners\n - 'email@example.com' in owners\n - 'email@example.com' in writers\n - 'email@example.com' in readers\n - starred = true\n - trashed = false\n - modifiedTime > '2024-01-01T00:00:00'\n - createdTime > '2024-01-01T00:00:00'\n - 'FOLDER_ID' in parents\n \n Combine with: and, or, not\n \n IMPORTANT:\n - Always include `trashed = false` unless the user explicitly asks for trash / deleted files.\n - Prefer `fullText contains` for topic/keyword folders (it searches name + indexed content).\n Use `name contains` when the intent is specifically filename-based.\n - Use RFC3339 timestamps for createdTime/modifiedTime. Prefer including a timezone, e.g. '2026-01-22T00:00:00Z'.\n \n DEFAULTS:\n - limit: 50\n - filename_format: {name}_{id}\n \n Query examples:\n - \"recent\" → trashed = false and modifiedTime > '2026-01-22T00:00:00Z'\n - \"shared\" → trashed = false and sharedWithMe = true\n - \"my-docs\" → trashed = false and 'me' in owners and mimeType = 'application/vnd.google-apps.document'\n - \"pdfs\" → trashed = false and mimeType = 'application/pdf'\n - \"documents\" → trashed = false and mimeType = 'application/vnd.google-apps.document'\n - \"spreadsheets\" → trashed = false and mimeType = 'application/vnd.google-apps.spreadsheet'\n - \"presentations\" → trashed = false and mimeType = 'application/vnd.google-apps.presentation'\n - \"images\" → trashed = false and mimeType contains 'image/'\n - \"starred\" → trashed = false and starred = true\n - \"college-docs\" → trashed = false and (fullText contains 'college' or name contains 'college') and mimeType = 'application/vnd.google-apps.document'\n - Guidance: \"created in the past week\" → use createdTime > '...'\n - Guidance: \"modified in the past week\" → use modifiedTime > '...'\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {name}, {date}, {created}, {mime_type}, {ext}\n - {name} is the basename (no extension)\n - {ext} is the extension (includes leading dot)\n - For documents: {name}_{id}\n - For dated files: {date}_{name}_{id}\n - For created-time focused queries: {created}_{name}_{id}\n - Default: {name}_{id}\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Notion Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferNotionQuery(name: string, guidance: string?) -> NotionQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Notion search query.\n \n Notion search is simple text-based. Extract key search terms from the folder name.\n Remove common words like \"my\", \"the\", \"all\", etc.\n \n DEFAULTS:\n - limit: 50\n - filename_format: {title}_{id}.md\n \n Query examples:\n - \"meeting-notes\" → meeting notes\n - \"project-docs\" → project docs\n - \"todo\" → todo\n - \"weekly-reports\" → weekly reports\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {title}, {type}, {date}, {created}\n - For notes: {title}_{id}.md\n - For dated pages: {date}_{title}_{id}.md\n - Default: {title}_{id}.md\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Slack Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferSlackQuery(name: string, guidance: string?) -> SlackQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Slack search query.\n \n Slack search supports various operators for finding messages across channels.\n The query should find relevant messages based on the folder name intent.\n \n SLACK SEARCH OPERATORS:\n - in:#channel-name - messages in a specific channel\n - in:@username - direct messages with a user\n - from:@username - messages from a specific user\n - from:me - messages you sent\n - to:me - messages sent to you (DMs)\n - has:link - messages containing links\n - has:reaction - messages with reactions\n - has:star - starred messages\n - has:pin - pinned messages\n - before:YYYY-MM-DD - messages before a date\n - after:YYYY-MM-DD - messages after a date\n - during:month - messages from a specific month (e.g., during:january)\n - on:YYYY-MM-DD - messages on a specific date\n - is:saved - saved messages\n \n QUERY STRATEGY:\n 1. For channel-specific queries → use in:#channel\n 2. For user-specific queries → use from:@user\n 3. For topic searches → use keywords directly\n 4. For time-based queries → use before:, after:, during:, on:\n 5. For content type → use has:link, has:reaction, etc.\n 6. Combine operators with spaces (implicit AND) or OR\n \n DEFAULTS:\n - limit: 50\n - filename_format: {date}_{channel}_{user}_{id}.txt\n \n Query examples:\n - \"engineering-updates\" → in:#engineering OR in:#engineering-updates\n - \"team-announcements\" → in:#announcements OR in:#general announcement\n - \"from-ceo\" → from:@ceo OR from:@founder\n - \"recent-links\" → has:link after:2026-01-01\n - \"pinned-messages\" → has:pin\n - \"design-feedback\" → in:#design feedback OR in:#design-reviews\n - \"standup-notes\" → standup OR \"daily standup\" OR in:#standup\n - \"customer-issues\" → in:#support OR in:#customer-success issue OR bug\n - \"product-discussions\" → in:#product OR product roadmap OR feature\n - \"onboarding\" → onboarding OR \"new hire\" OR welcome\n - \"decisions\" → decision OR decided OR \"we will\" OR approved\n - \"action-items\" → \"action item\" OR TODO OR \"follow up\" OR assigned\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {channel}, {user}, {date}, {text}\n - For channel messages: {date}_{channel}_{user}_{id}.txt\n - For user messages: {date}_{user}_{channel}_{id}.txt\n - Default: {date}_{channel}_{user}_{id}.txt\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Linear Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferLinearQuery(name: string, guidance: string?) -> LinearQueryResult {\n client AnthropicClient\n prompt #\"\n Convert natural language into a Linear search query.\n \n CORE PRINCIPLE: Text search is powerful - use it liberally. Only add filters when intent is unambiguous.\n \n SYNTAX:\n - Plain text: searches title + description (very flexible, use this a lot!)\n - assignee:name | assignee:me - who it's assigned to\n - no:assignee - unassigned issues\n - state:backlog | state:todo | state:\"in progress\" | state:done | state:canceled\n - priority:1 (urgent) | priority:2 (high) | priority:3 (medium) | priority:4 (low) \n - label:name - filter by label\n - no:label - unlabeled issues\n - team:KEY - filter by team key\n - project:name - filter by project name\n - creator:name - who created it\n - cycle:current | cycle:next | cycle:previous - filter by cycle\n - estimate:N - filter by point estimate\n - OR - combine alternatives (use generously!)\n \n STRATEGY - Think about what the user ACTUALLY wants:\n \n 1. TOPIC/FEATURE searches → mostly text search\n \"checkout bugs\" → checkout bug\n \"api performance\" → api performance\n \"mobile crashes\" → mobile crash\n \"auth issues\" → auth authentication login\n \n 2. PERSON searches → assignee filter + text backup\n \"john's tickets\" → assignee:john OR john\n \"sarah jones work\" → assignee:sarah OR assignee:jones OR sarah OR jones\n \"assigned to mike\" → assignee:mike\n \"unassigned issues\" → no:assignee\n \n 3. STATUS/PRIORITY searches → filters\n \"my open issues\" → assignee:me state:todo OR state:\"in progress\"\n \"urgent bugs\" → priority:1 bug\n \"blocked tickets\" → blocked OR blocking\n \"stuff in review\" → review\n \"in progress issues\" → state:\"in progress\"\n \n 4. PROJECT/CYCLE searches → project or cycle filter\n \"acme project\" → project:acme\n \"current sprint\" → cycle:current\n \"next sprint issues\" → cycle:next\n \"q1 roadmap\" → q1 roadmap\n \"enterprise features\" → enterprise\n \n 5. VAGUE/BROAD searches → text search with synonyms\n \"tech debt\" → \"tech debt\" OR refactor OR cleanup\n \"security\" → security vulnerability\n \"onboarding\" → onboarding OR \"getting started\" OR setup\n \n KEY RULES:\n - When in doubt, use text search - it's flexible\n - Use OR generously to catch variations \n - For multi-word names: assignee:first OR assignee:last (never the full name)\n - Add synonyms for common concepts\n - Don't over-filter - better to return more results than miss things\n \n SEARCH TYPE: \"issues\" (default) or \"projects\" (only if explicitly asking for projects list)\n \n DEFAULTS:\n - limit: 50\n - filename_format: {identifier}_{title}.md\n \n CRITICAL: filename_format MUST contain placeholders like {identifier}, {title}, etc.\n NEVER return a literal filename like \"test.md\" or the folder name as the format.\n The format is a TEMPLATE applied to every result, so it MUST use placeholders.\n \n Folder/file name: {{ name }}\n {% if guidance %}\n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// GitHub Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGitHubQuery(name: string, guidance: string?) -> GitHubQueryResult {\n client AnthropicClient\n prompt #\"\n Convert natural language into a GitHub query. Think about what an agent would need.\n \n ═══════════════════════════════════════════════════════════════════════════════\n SEARCH TYPES & QUERY FORMATS\n ═══════════════════════════════════════════════════════════════════════════════\n \n 1. REPOS (search_type: \"repos\")\n Listing: list:org/ORGNAME or list:user/USERNAME [type:private|public]\n Search: language:go stars:>100 topic:ml org:company\n \n Examples:\n - \"beam-cloud repos\" → list:org/beam-cloud\n - \"private repos in acme\" → list:org/acme type:private\n - \"popular go projects\" → language:go stars:>1000\n \n 2. ISSUES (search_type: \"issues\") \n Query: repo:owner/name [is:open|closed] [label:X] [assignee:X] [author:X]\n \n Examples:\n - \"open bugs in beta9\" → repo:beam-cloud/beta9 is:open label:bug\n - \"my issues\" → assignee:@me is:open\n - \"security issues in react\" → repo:facebook/react label:security\n \n 3. PULL REQUESTS (search_type: \"prs\")\n Query: repo:owner/name [is:open|closed|merged] [author:X] [review:approved]\n \n Examples:\n - \"open PRs in beta9\" → repo:beam-cloud/beta9 is:open\n - \"merged PRs this week\" → repo:owner/repo is:merged merged:>2026-01-26\n - \"PRs needing review\" → repo:owner/repo is:open review:none\n \n 4. COMMITS (search_type: \"commits\")\n Query: repo:owner/name [branch:X] [author:X] [since:DATE] [path:X]\n \n Examples:\n - \"recent commits in beta9\" → repo:beam-cloud/beta9\n - \"commits to main this week\" → repo:owner/repo branch:main since:2026-01-26\n - \"john's commits\" → repo:owner/repo author:john\n - \"changes to src/api\" → repo:owner/repo path:src/api\n \n 5. RELEASES (search_type: \"releases\")\n Query: repo:owner/name [latest] [prerelease:true|false]\n \n Examples:\n - \"beta9 releases\" → repo:beam-cloud/beta9\n - \"latest release of kubernetes\" → repo:kubernetes/kubernetes latest\n - \"stable releases only\" → repo:owner/repo prerelease:false\n \n 6. WORKFLOW RUNS (search_type: \"workflows\")\n Query: repo:owner/name [status:success|failure|in_progress] [branch:X] [event:push|pr]\n \n Examples:\n - \"failed CI in beta9\" → repo:beam-cloud/beta9 status:failure\n - \"recent builds\" → repo:owner/repo\n - \"PR checks\" → repo:owner/repo event:pull_request\n \n 7. FILES (search_type: \"files\")\n Query: repo:owner/name path:PATH [ref:branch|tag|sha]\n \n Examples:\n - \"beta9 readme\" → repo:beam-cloud/beta9 path:README.md\n - \"package.json in react\" → repo:facebook/react path:package.json\n - \"config files\" → repo:owner/repo path:*.yaml ref:main\n - \"src directory\" → repo:owner/repo path:src/\n \n 8. BRANCHES (search_type: \"branches\")\n Query: repo:owner/name [protected:true|false]\n \n Examples:\n - \"beta9 branches\" → repo:beam-cloud/beta9\n - \"protected branches\" → repo:owner/repo protected:true\n \n ═══════════════════════════════════════════════════════════════════════════════\n CONTENT TYPES (how to format the content)\n ═══════════════════════════════════════════════════════════════════════════════\n \n - \"markdown\" (default) - Rich formatted content, best for agents to read\n - \"diff\" - Code changes (for PRs/commits)\n - \"json\" - Structured metadata\n - \"raw\" - Unprocessed file content (for files)\n \n ═══════════════════════════════════════════════════════════════════════════════\n FILENAME FORMATS (use appropriate placeholders)\n ═══════════════════════════════════════════════════════════════════════════════\n \n Repos: {name}.md or {full_name}.json\n Issues: {number}_{title}.md\n PRs: {number}_{title}.md or {number}_{title}_diff.md\n Commits: {sha_short}_{message}.md\n Releases: {tag}_{name}.md\n Workflows: {id}_{name}_{status}.md\n Files: {path} (preserve original path/name)\n Branches: {name}.md\n \n ═══════════════════════════════════════════════════════════════════════════════\n SMART DETECTION\n ═══════════════════════════════════════════════════════════════════════════════\n \n Detect intent from keywords:\n - \"repos/repositories/projects\" in org/user → repos listing\n - \"prs/pulls/pull requests/merge requests\" → prs\n - \"issues/bugs/tickets/tasks\" → issues\n - \"commits/history/changes/log\" → commits \n - \"releases/versions/tags\" → releases\n - \"ci/cd/builds/workflows/actions/checks\" → workflows\n - \"readme/config/file/code\" → files\n - \"branches\" → branches\n - \"diff/patch/changes\" → content_type: diff\n \n Folder/file name: {{ name }}\n {% if guidance %}\n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Tests\n// -----------------------------------------------------------------------------\n\ntest TestGitHubOpenPRs {\n functions [InferGitHubQuery]\n args {\n name \"open-prs-beta9\"\n guidance null\n }\n}\n\ntest TestGitHubIssues {\n functions [InferGitHubQuery]\n args {\n name \"beta9-bugs\"\n guidance \"show open bugs in the beta9 repo\"\n }\n}\n\ntest TestGitHubRepos {\n functions [InferGitHubQuery]\n args {\n name \"go-cli-tools\"\n guidance null\n }\n}\n\ntest TestGmailUnread {\n functions [InferGmailQuery]\n args {\n name \"unread-emails\"\n guidance null\n }\n}\n\ntest TestGmailFromPerson {\n functions [InferGmailQuery]\n args {\n name \"from-eli\"\n guidance null\n }\n}\n\ntest TestGmailWithGuidance {\n functions [InferGmailQuery]\n args {\n name \"important-emails\"\n guidance \"Only from the last 7 days, max 100 results\"\n }\n}\n\ntest TestGmailInvoices {\n functions [InferGmailQuery]\n args {\n name \"invoices\"\n guidance null\n }\n}\n\ntest TestGmailMeetingNotes {\n functions [InferGmailQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestGmailFlightConfirmation {\n functions [InferGmailQuery]\n args {\n name \"flight-confirmations\"\n guidance null\n }\n}\n\ntest TestGmailReceipts {\n functions [InferGmailQuery]\n args {\n name \"receipts\"\n guidance null\n }\n}\n\ntest TestGmailProjectAlpha {\n functions [InferGmailQuery]\n args {\n name \"project-alpha-emails\"\n guidance null\n }\n}\n\ntest TestGmailShippingUpdates {\n functions [InferGmailQuery]\n args {\n name \"shipping-updates\"\n guidance null\n }\n}\n\ntest TestGmailHealthWithGuidance {\n functions [InferGmailQuery]\n args {\n name \"Medical Documents\"\n guidance \"anything from: summit health, nyu, anything health related basically, mychart\"\n }\n}\n\ntest TestGmailMultiWordSender {\n functions [InferGmailQuery]\n args {\n name \"acme-corp-emails\"\n guidance \"from acme corp and related companies\"\n }\n}\n\ntest TestGmailFuzzySenderNike {\n functions [InferGmailQuery]\n args {\n name \"nike-emails\"\n guidance \"from nike\"\n }\n}\n\ntest TestGmailFuzzySenderAmazon {\n functions [InferGmailQuery]\n args {\n name \"amazon-orders\"\n guidance null\n }\n}\n\ntest TestGDriveShared {\n functions [InferGDriveQuery]\n args {\n name \"shared-with-me\"\n guidance null\n }\n}\n\ntest TestGDrivePdfs {\n functions [InferGDriveQuery]\n args {\n name \"pdfs\"\n guidance null\n }\n}\n\ntest TestGDriveCollegeDocs {\n functions [InferGDriveQuery]\n args {\n name \"college docs\"\n guidance null\n }\n}\n\ntest TestGDriveRecentDocsWithGuidance {\n functions [InferGDriveQuery]\n args {\n name \"recent-docs\"\n guidance \"any docs created in the past week\"\n }\n}\n\ntest TestGDriveOwnedByMe {\n functions [InferGDriveQuery]\n args {\n name \"my docs\"\n guidance null\n }\n}\n\ntest TestNotionMeetings {\n functions [InferNotionQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestEvaluateGmailResults_NoResults {\n functions [EvaluateGmailQueryResults]\n args {\n original_guidance \"anything from summit health, nyu, mychart, health related\"\n generated_query \"from:summit health OR from:nyu OR from:mychart\"\n result_count 0\n sample_results \"[]\"\n }\n}\n\ntest TestEvaluateGmailResults_GoodResults {\n functions [EvaluateGmailQueryResults]\n args {\n original_guidance \"invoices from stripe\"\n generated_query \"invoice from:stripe\"\n result_count 15\n sample_results \"[{\\\"from\\\":\\\"Stripe\\\",\\\"subject\\\":\\\"Your invoice from Acme Corp\\\",\\\"snippet\\\":\\\"Invoice #12345...\\\"},{\\\"from\\\":\\\"Stripe Billing\\\",\\\"subject\\\":\\\"Invoice for January\\\",\\\"snippet\\\":\\\"Amount due: $99\\\"}]\"\n }\n}\n\ntest TestSlackEngineering {\n functions [InferSlackQuery]\n args {\n name \"engineering-updates\"\n guidance null\n }\n}\n\ntest TestSlackFromCEO {\n functions [InferSlackQuery]\n args {\n name \"from-ceo\"\n guidance \"messages from our CEO\"\n }\n}\n\ntest TestSlackRecentLinks {\n functions [InferSlackQuery]\n args {\n name \"recent-links\"\n guidance null\n }\n}\n\ntest TestSlackStandups {\n functions [InferSlackQuery]\n args {\n name \"standup-notes\"\n guidance null\n }\n}\n\ntest TestSlackDecisions {\n functions [InferSlackQuery]\n args {\n name \"team-decisions\"\n guidance \"important decisions made by the team\"\n }\n}\n\ntest TestLinearMyIssues {\n functions [InferLinearQuery]\n args {\n name \"my-issues\"\n guidance null\n }\n}\n\ntest TestLinearMyBugs {\n functions [InferLinearQuery]\n args {\n name \"my-bugs\"\n guidance \"bugs assigned to me\"\n }\n}\n\ntest TestLinearUrgent {\n functions [InferLinearQuery]\n args {\n name \"urgent-issues\"\n guidance null\n }\n}\n\ntest TestLinearTeamBacklog {\n functions [InferLinearQuery]\n args {\n name \"eng-backlog\"\n guidance \"engineering team backlog\"\n }\n}\n\ntest TestLinearInProgress {\n functions [InferLinearQuery]\n args {\n name \"in-progress\"\n guidance null\n }\n}\n\ntest TestPostHogActiveFlags {\n functions [InferPostHogQuery]\n args {\n name \"active-feature-flags\"\n guidance null\n }\n}\n\ntest TestPostHogPageviewEvents {\n functions [InferPostHogQuery]\n args {\n name \"pageview-events\"\n guidance null\n }\n}\n\ntest TestPostHogRetentionInsights {\n functions [InferPostHogQuery]\n args {\n name \"retention-insights\"\n guidance null\n }\n}\n\ntest TestPostHogBetaUsers {\n functions [InferPostHogQuery]\n args {\n name \"beta-users\"\n guidance null\n }\n}\n\ntest TestPostHogErrorEvents {\n functions [InferPostHogQuery]\n args {\n name \"error-events\"\n guidance \"exception and error events from the past week\"\n }\n}\n\ntest TestPostHogSignupFunnel {\n functions [InferPostHogQuery]\n args {\n name \"signup-funnel\"\n guidance null\n }\n}\n\n// -----------------------------------------------------------------------------\n// Confluence Query Inference\n// -----------------------------------------------------------------------------\n\nclass ConfluenceQueryResult {\n cql_query string @description(\"Confluence Query Language (CQL) query string\")\n content_type string @description(\"Content type filter: 'page', 'blogpost', or 'all'. Default: 'all'\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {title}, {space}, {date}, {type}. Example: {space}_{title}_{id}.md\")\n}\n\nfunction InferConfluenceQuery(name: string, guidance: string?) -> ConfluenceQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Confluence CQL (Confluence Query Language) query.\n\n CQL SYNTAX:\n - type=page — pages only\n - type=blogpost — blog posts only\n - space=KEY — filter by space key (e.g., space=DEV, space=ENG)\n - title~\"text\" — title contains text (fuzzy match)\n - text~\"text\" — full-text search (searches title + body)\n - label=\"name\" — filter by label\n - creator=\"username\" — filter by creator\n - contributor=\"username\" — filter by contributor\n - lastModified>now(\"-7d\") — modified in the last 7 days\n - lastModified>now(\"-30d\") — modified in the last 30 days\n - lastModified>\"2025-01-01\" — modified after a specific date\n - created>now(\"-7d\") — created in the last 7 days\n - ancestor=12345 — pages under a specific parent\n - AND, OR — combine conditions\n - ORDER BY lastModified DESC, title ASC — sorting\n\n QUERY STRATEGY:\n 1. For topic searches → use text~ for full-text search\n 2. For space-specific → use space=KEY\n 3. For recent content → use lastModified>now(\"-Nd\")\n 4. For labeled content → use label=\"name\"\n 5. For specific content types → use type=page or type=blogpost\n 6. Combine operators with AND/OR\n\n DEFAULTS:\n - content_type: \"all\"\n - limit: 50\n - filename_format: {space}_{title}_{id}.md\n\n Query examples:\n - \"recent-pages\" → lastModified>now(\"-7d\") ORDER BY lastModified DESC, content_type: \"page\"\n - \"engineering-docs\" → (space=ENG OR text~\"engineering\") AND type=page ORDER BY lastModified DESC\n - \"kubernetes-pages\" → text~\"kubernetes\" AND type=page ORDER BY lastModified DESC\n - \"dev-blog-posts\" → space=DEV AND type=blogpost ORDER BY lastModified DESC\n - \"important-docs\" → label=\"important\" ORDER BY lastModified DESC\n - \"api-documentation\" → text~\"api\" AND (label=\"documentation\" OR title~\"api\") ORDER BY lastModified DESC\n - \"onboarding\" → text~\"onboarding\" OR title~\"onboarding\" ORDER BY lastModified DESC\n - \"architecture-decisions\" → text~\"architecture\" OR label=\"adr\" ORDER BY lastModified DESC\n - \"meeting-notes\" → title~\"meeting\" OR label=\"meeting-notes\" ORDER BY lastModified DESC\n - \"recent-blog-posts\" → type=blogpost AND lastModified>now(\"-30d\") ORDER BY lastModified DESC\n - \"project-alpha\" → text~\"project alpha\" ORDER BY lastModified DESC\n\n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {title}, {space}, {date}, {type}\n - For space-specific: {title}_{id}.md\n - For cross-space: {space}_{title}_{id}.md\n - For dated content: {date}_{title}_{id}.md\n - Default: {space}_{title}_{id}.md\n\n Folder/file name: {{ name }}\n {% if guidance %}\n\n Additional guidance from user: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Web Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferWebQuery(name: string, guidance: string?) -> WebQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder name into a web query. Two modes:\n\n web_mode=map — crawl a specific website. web_query must be a valid https:// URL.\n web_mode=search — web search. web_query is a search query string (not a URL).\n\n Use web_mode=map when guidance contains a URL or the name implies a specific site.\n Use web_mode=search when the name describes a topic, question, or information need.\n\n EXAMPLES:\n - \"hennessy-cocktails\" (guidance: \"https://www.hennessy.com/en-us/cocktails\")\n → web_mode: \"map\", web_query: \"https://www.hennessy.com/en-us/cocktails\", include_paths: [\"/en-us/cocktails/*\"], limit: 100\n - \"react-docs\"\n → web_mode: \"map\", web_query: \"https://react.dev/reference\", include_paths: [\"/reference/*\"], limit: 100\n - \"python-tutorial\"\n → web_mode: \"map\", web_query: \"https://docs.python.org/3/tutorial/\", include_paths: [\"/3/tutorial/*\"], limit: 50\n - \"latest AI news\"\n → web_mode: \"search\", web_query: \"latest AI news 2025\", include_paths: [], limit: 20\n - \"best rust web frameworks\"\n → web_mode: \"search\", web_query: \"best rust web frameworks comparison\", include_paths: [], limit: 15\n - \"climate change research papers\"\n → web_mode: \"search\", web_query: \"climate change research papers 2025\", include_paths: [], limit: 20\n\n Defaults: web_mode=map, limit=100, filename_format={title}_{id}.md\n\n Folder name: {{ name }}\n {% if guidance %}\n User guidance: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Web Query Tests\n// -----------------------------------------------------------------------------\n\ntest TestWebQueryWithGuidance {\n functions [InferWebQuery]\n args {\n name \"hennessy-cocktails\"\n guidance \"https://www.hennessy.com/en-us/cocktails?page=1\"\n }\n}\n\ntest TestWebQueryDocs {\n functions [InferWebQuery]\n args {\n name \"react-docs\"\n guidance null\n }\n}\n\ntest TestWebQueryBlog {\n functions [InferWebQuery]\n args {\n name \"openai-blog\"\n guidance null\n }\n}\n\ntest TestWebQuerySearch {\n functions [InferWebQuery]\n args {\n name \"latest AI news\"\n guidance null\n }\n}\n\ntest TestWebQuerySearchTopic {\n functions [InferWebQuery]\n args {\n name \"best rust web frameworks\"\n guidance null\n }\n}\n\n// -----------------------------------------------------------------------------\n// Confluence Query Tests\n// -----------------------------------------------------------------------------\n\ntest TestConfluenceRecentPages {\n functions [InferConfluenceQuery]\n args {\n name \"recent-pages\"\n guidance null\n }\n}\n\ntest TestConfluenceEngDocs {\n functions [InferConfluenceQuery]\n args {\n name \"engineering-docs\"\n guidance \"documentation from the engineering space\"\n }\n}\n\ntest TestConfluenceKubernetes {\n functions [InferConfluenceQuery]\n args {\n name \"kubernetes-pages\"\n guidance null\n }\n}\n\ntest TestConfluenceMeetingNotes {\n functions [InferConfluenceQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestConfluenceArchitectureDecisions {\n functions [InferConfluenceQuery]\n args {\n name \"architecture-decisions\"\n guidance \"ADRs and architecture decision records\"\n }\n}\n",
+ "generators.baml": "generator target {\n output_type \"go\"\n output_dir \"../\"\n version \"0.218.1\"\n\n // 'baml-cli generate' will run this after generating go code\n // This command will be run from within $output_dir/baml_client\n on_generate \"bash \\\"$(git rev-parse --show-toplevel)/bin/patch-baml-runtime.sh\\\" && gofmt -w . && goimports -w .\"\n\n // Go packages name as specified in go.mod\n // We need this to generate correct imports in the generated baml_client\n client_package_name \"github.com/beam-cloud/airstore/pkg/sources/queries\"\n}\n",
+ "smart_queries.baml": "// =============================================================================\n// Smart Query BAML Functions\n// Infer source-specific queries from folder/file names\n// =============================================================================\n\nclass GmailQueryResult {\n gmail_query string @description(\"Gmail search query using Gmail's operators\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {date}, {from}, {subject}. Example: {date}_{from}_{subject}_{id}.txt\")\n}\n\nclass GDriveQueryResult {\n gdrive_query string @description(\"Google Drive search query\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {name}, {date}. Example: {name}_{id}\")\n}\n\nclass NotionQueryResult {\n notion_query string @description(\"Notion search query text\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {title}, {date}. Example: {title}_{id}.md\")\n}\n\nclass GitHubQueryResult {\n github_query string @description(\"GitHub query - format depends on search_type\")\n search_type string @description(\"Type: repos, issues, prs, commits, releases, workflows, files, branches\")\n content_type string @description(\"Format: markdown (default), diff (PRs/commits), json (metadata), raw (files)\")\n limit int @description(\"Max results (default: 50)\")\n filename_format string @description(\"Filename template with placeholders like {number}, {title}, {sha}, {name}, {path}\")\n}\n\nclass SlackQueryResult {\n slack_query string @description(\"Slack search query using Slack's search syntax\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {channel}, {user}, {date}, {text}. Example: {date}_{channel}_{user}_{id}.txt\")\n}\n\nclass LinearQueryResult {\n linear_query string @description(\"Linear issue filter query using Linear's filter syntax\")\n search_type string @description(\"Type of search: 'issues' or 'projects'. Default: 'issues'\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {identifier}, {title}, {state}, {assignee}, {team}, {priority}, {date}. Example: {identifier}_{title}.md\")\n}\n\nclass PostHogQueryResult {\n posthog_query string @description(\"Search query string for PostHog API\")\n search_type string @description(\"Type of search: 'events', 'feature-flags', 'insights', 'cohorts'\")\n project_id int @description(\"PostHog project ID to search in. Use 0 to auto-select the first available project.\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Placeholders vary by search_type.\")\n}\n\nclass WebQueryResult {\n web_mode string @description(\"'map' to crawl a specific site, 'search' to find pages via web search. Default: map\")\n web_query string @description(\"For web_mode=map: the URL to crawl. For web_mode=search: the search query string.\")\n include_paths string[] @description(\"URL path patterns to include when web_mode=map (e.g. /cocktails/*). Empty array = all paths. Ignored for search.\")\n limit int @description(\"Max pages to discover (default: 100)\")\n filename_format string @description(\"Template for result filenames. Use {title}, {path}, {id}, {date}. Example: {title}_{id}.md\")\n}\n\n// -----------------------------------------------------------------------------\n// PostHog Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferPostHogQuery(name: string, guidance: string?) -> PostHogQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a PostHog query.\n\n SEARCH TYPES:\n 1. feature-flags — Feature flags (toggles). Use when name contains: flag, toggle, feature, rollout, experiment\n 2. insights — Saved queries, funnels, retention, trends. Use when name contains: insight, funnel, retention, trend, dashboard, chart, analytics\n 3. cohorts — User segments/groups. Use when name contains: cohort, segment, group, users, audience\n 4. events — Raw event data (default). Use when name contains: event, pageview, click, action, or nothing specific\n\n POSTHOG STANDARD EVENTS (start with $):\n $pageview, $pageleave, $autocapture, $screen, $exception, $rageclick, $web_vitals\n\n QUERY FORMAT:\n - For events: the event name to filter on (e.g., \"$pageview\", \"signup\", \"purchase\"). Empty string = all recent events.\n - For feature-flags/insights/cohorts: a search term to filter by name/key (e.g., \"beta\", \"onboarding\", \"retention\").\n\n ACTIVE FLAGS:\n If the name implies \"active\" or \"enabled\" flags (e.g., \"active-feature-flags\", \"enabled-flags\"), set search_type to \"feature-flags\" and use a broad query (empty or relevant keyword). The provider will filter for active=true.\n\n DEFAULTS:\n - project_id: 0 (auto-select first project)\n - limit: 50\n\n FILENAME FORMATS by search_type:\n - events: {date}_{event}_{id}.json\n - feature-flags: {key}_{id}.json\n - insights: {name}_{id}.json\n - cohorts: {name}_{id}.json\n\n Query examples:\n - \"active-feature-flags\" → search_type: feature-flags, query: \"\" (all flags, provider filters active)\n - \"pageview-events\" → search_type: events, query: \"$pageview\"\n - \"error-events\" → search_type: events, query: \"$exception\"\n - \"retention-insights\" → search_type: insights, query: \"retention\"\n - \"beta-users\" → search_type: cohorts, query: \"beta\"\n - \"signup-funnel\" → search_type: insights, query: \"signup funnel\"\n - \"recent-events\" → search_type: events, query: \"\"\n - \"experiment-flags\" → search_type: feature-flags, query: \"experiment\"\n - \"rage-clicks\" → search_type: events, query: \"$rageclick\"\n - \"web-vitals\" → search_type: events, query: \"$web_vitals\"\n - \"onboarding-cohort\" → search_type: cohorts, query: \"onboarding\"\n\n Folder/file name: {{ name }}\n {% if guidance %}\n\n Additional guidance from user: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Gmail Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGmailQuery(name: string, guidance: string?) -> GmailQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Gmail search query.\n \n IMPORTANT: Gmail search is FUZZY by default!\n - Bare words (no operator) search across subject, body, AND sender name/email\n - Use bare words for most searches - they match everywhere!\n - Gmail automatically handles plurals, common misspellings, and related terms\n \n CRITICAL: PREFER BARE WORDS OVER from: FOR FUZZY SENDER MATCHING!\n - from:company only matches if that word appears exactly in sender name/email\n - Bare words match sender, subject, AND body - much more flexible!\n - Many companies use subdomains like noreply@mail.company.com - bare words catch these better\n - ONLY use from: when you need STRICT sender matching (rare)\n \n CRITICAL SYNTAX RULES:\n - Multi-word operator values MUST be quoted: from:\"summit health\" NOT from:summit health\n - Bare multi-word phrases should also be quoted: \"lab results\" NOT lab results\n - Unquoted spaces mean AND: lab results = lab AND results (both must appear)\n - Use OR to match alternatives: mychart OR \"nyu langone\" OR \"summit health\"\n \n Gmail search operators (use sparingly - bare words are usually better!):\n - from:\"name or email\" - STRICT sender match (avoid for fuzzy matching!)\n - subject:\"phrase\" - matches subject line only\n - has:attachment, filename:pdf\n - is:unread, is:starred, is:important\n - newer_than:1d, newer_than:7d, older_than:30d\n \n QUERY STRATEGY:\n 1. For company/brand emails → USE BARE WORDS: nike OR adidas OR \"north face\"\n (This finds emails FROM them, ABOUT them, or MENTIONING them - much better coverage!)\n 2. For topics → bare words or quoted phrases: \"lab results\" OR prescription OR appointment\n 3. For people → from:firstname works but bare word also good: john OR from:john\n 4. For mixed (topic + senders) → all bare words: mychart OR nyu OR \"summit health\" OR prescription\n 5. For health/medical → medical OR health OR doctor OR appointment OR mychart OR prescription\n 6. When in doubt → ADD MORE ORs with related terms, synonyms, and variations!\n \n DEFAULTS:\n - limit: 50\n - filename_format: {date}_{from}_{subject}_{id}.txt\n \n Query examples:\n - \"invoices\" → invoice OR receipt OR bill OR statement\n - \"stripe-invoices\" → stripe invoice OR stripe receipt OR stripe payment\n - \"receipts\" → receipt OR invoice OR order OR confirmation OR purchase\n - \"meeting-notes\" → meeting OR agenda OR \"calendar invite\" OR standup OR sync\n - \"flight-confirmation\" → flight OR itinerary OR \"boarding pass\" OR airline OR booking\n - \"health-emails\" → medical OR health OR doctor OR appointment OR mychart OR prescription OR lab\n - \"from-john\" → from:john (explicit \"from\" in name = use the operator)\n - \"emails-from-acme-corp\" → from:\"acme corp\" (explicit \"from\" + multi-word = quoted operator)\n - \"unread\" → is:unread\n - \"project-alpha\" → \"project alpha\" (quoted phrase)\n - \"shipping\" → shipping OR tracking OR delivery OR fedex OR ups OR usps OR package\n - \"recent\" → newer_than:7d\n - \"amazon-orders\" → amazon order OR amazon shipment OR amazon delivery OR amazon purchase\n - \"uber-receipts\" → uber receipt OR uber ride OR lyft receipt OR lyft ride\n \n Guidance examples:\n - \"from nike\" → nike (bare word for fuzzy matching - catches nike.com, news@nike.com, etc.)\n - \"anything from summit health, nyu, mychart, health related\" \n → mychart OR nyu OR \"summit health\" OR medical OR appointment OR prescription OR lab\n (Use bare words with lots of ORs - they match sender names AND content!)\n - \"strictly only emails where the sender is john@company.com\" → from:john@company.com\n (Use from: ONLY when user explicitly wants strict sender matching)\n \n Filename format: {date}_{from}_{subject}_{id}.txt\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Gmail Query Evaluation and Refinement\n// -----------------------------------------------------------------------------\n\nclass GmailQueryEvaluation {\n is_satisfactory bool @description(\"True if results match user intent well enough\")\n refined_query string? @description(\"Improved Gmail query if not satisfactory, null if satisfactory\")\n reasoning string @description(\"Brief explanation of evaluation (1-2 sentences)\")\n}\n\nfunction EvaluateGmailQueryResults(\n original_guidance: string,\n generated_query: string,\n result_count: int,\n sample_results: string\n) -> GmailQueryEvaluation {\n client AnthropicClient\n prompt #\"\n You are a STRICT evaluator checking if Gmail search results actually match what the user asked for.\n Be critical - if results don't clearly match the user's intent, suggest a better query.\n \n USER'S REQUEST: {{ original_guidance }}\n \n CURRENT QUERY: {{ generated_query }}\n \n RESULTS: {{ result_count }} emails found\n \n SAMPLE RESULTS: {{ sample_results }}\n \n STRICT EVALUATION CHECKLIST:\n \n 1. ZERO RESULTS = AUTOMATIC FAIL\n - If result_count is 0, the query is broken. Mark as NOT satisfactory.\n - Common causes:\n * Unquoted multi-word values: from:summit health → from:\"summit health\"\n * Using from: when bare word would match better (from:bonobos might miss noreply@e.bonobos.com)\n \n 2. CHECK EACH THING THE USER MENTIONED:\n - If user mentioned specific senders/companies (e.g., \"bonobos\", \"amazon\", \"stripe\"):\n → Are emails FROM or ABOUT these in the results?\n → Note: bare words like \"bonobos\" are GOOD - they match sender, subject, AND body\n → If from: operator fails, suggest using bare word instead for fuzzy matching\n - If user mentioned topics (e.g., \"health\", \"medical\", \"appointments\"):\n → Do the subjects/snippets actually contain these words?\n → Generic results that don't mention the topics = NOT satisfactory.\n \n 3. LOOK FOR OBVIOUS MISMATCHES:\n - Results about completely unrelated topics = NOT satisfactory\n - Missing the main thing the user asked for = NOT satisfactory\n \n 4. QUERY IMPROVEMENTS TO SUGGEST:\n - from:bonobos (0 results) → bonobos (bare word matches subdomains like e.bonobos.com)\n - from:summit health → from:\"summit health\" (MUST quote multi-word)\n - Using from: when company uses weird email domains → use bare word instead\n - Not enough OR alternatives for fuzzy matching\n \n BE HARSH: If you're unsure whether results match, lean toward NOT satisfactory and suggest improvements.\n Only mark satisfactory if the results CLEARLY match what the user asked for.\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Google Drive Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGDriveQuery(name: string, guidance: string?) -> GDriveQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Google Drive files.list query (the `q` parameter).\n \n Output MUST be ONLY the query expression (no `q=`, no URL encoding).\n It MUST be valid Drive query syntax.\n \n Drive query syntax:\n - name contains 'text'\n - fullText contains 'text'\n - mimeType = 'application/pdf'\n - mimeType = 'application/vnd.google-apps.document'\n - mimeType = 'application/vnd.google-apps.spreadsheet'\n - mimeType = 'application/vnd.google-apps.presentation'\n - mimeType contains 'image'\n - sharedWithMe = true\n - 'me' in owners\n - 'email@example.com' in owners\n - 'email@example.com' in writers\n - 'email@example.com' in readers\n - starred = true\n - trashed = false\n - modifiedTime > '2024-01-01T00:00:00'\n - createdTime > '2024-01-01T00:00:00'\n - 'FOLDER_ID' in parents\n \n Combine with: and, or, not\n \n IMPORTANT:\n - Always include `trashed = false` unless the user explicitly asks for trash / deleted files.\n - Prefer `fullText contains` for topic/keyword folders (it searches name + indexed content).\n Use `name contains` when the intent is specifically filename-based.\n - Use RFC3339 timestamps for createdTime/modifiedTime. Prefer including a timezone, e.g. '2026-01-22T00:00:00Z'.\n \n DEFAULTS:\n - limit: 50\n - filename_format: {name}_{id}\n \n Query examples:\n - \"recent\" → trashed = false and modifiedTime > '2026-01-22T00:00:00Z'\n - \"shared\" → trashed = false and sharedWithMe = true\n - \"my-docs\" → trashed = false and 'me' in owners and mimeType = 'application/vnd.google-apps.document'\n - \"pdfs\" → trashed = false and mimeType = 'application/pdf'\n - \"documents\" → trashed = false and mimeType = 'application/vnd.google-apps.document'\n - \"spreadsheets\" → trashed = false and mimeType = 'application/vnd.google-apps.spreadsheet'\n - \"presentations\" → trashed = false and mimeType = 'application/vnd.google-apps.presentation'\n - \"images\" → trashed = false and mimeType contains 'image/'\n - \"starred\" → trashed = false and starred = true\n - \"college-docs\" → trashed = false and (fullText contains 'college' or name contains 'college') and mimeType = 'application/vnd.google-apps.document'\n - Guidance: \"created in the past week\" → use createdTime > '...'\n - Guidance: \"modified in the past week\" → use modifiedTime > '...'\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {name}, {date}, {created}, {mime_type}, {ext}\n - {name} is the basename (no extension)\n - {ext} is the extension (includes leading dot)\n - For documents: {name}_{id}\n - For dated files: {date}_{name}_{id}\n - For created-time focused queries: {created}_{name}_{id}\n - Default: {name}_{id}\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Notion Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferNotionQuery(name: string, guidance: string?) -> NotionQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Notion search query.\n \n Notion search is simple text-based. Extract key search terms from the folder name.\n Remove common words like \"my\", \"the\", \"all\", etc.\n \n DEFAULTS:\n - limit: 50\n - filename_format: {title}_{id}.md\n \n Query examples:\n - \"meeting-notes\" → meeting notes\n - \"project-docs\" → project docs\n - \"todo\" → todo\n - \"weekly-reports\" → weekly reports\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {title}, {type}, {date}, {created}\n - For notes: {title}_{id}.md\n - For dated pages: {date}_{title}_{id}.md\n - Default: {title}_{id}.md\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Slack Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferSlackQuery(name: string, guidance: string?) -> SlackQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Slack search query.\n \n Slack search supports various operators for finding messages across channels.\n The query should find relevant messages based on the folder name intent.\n \n SLACK SEARCH OPERATORS:\n - in:#channel-name - messages in a specific channel\n - in:@username - direct messages with a user\n - from:@username - messages from a specific user\n - from:me - messages you sent\n - to:me - messages sent to you (DMs)\n - has:link - messages containing links\n - has:reaction - messages with reactions\n - has:star - starred messages\n - has:pin - pinned messages\n - before:YYYY-MM-DD - messages before a date\n - after:YYYY-MM-DD - messages after a date\n - during:month - messages from a specific month (e.g., during:january)\n - on:YYYY-MM-DD - messages on a specific date\n - is:saved - saved messages\n \n QUERY STRATEGY:\n 1. For channel-specific queries → use in:#channel\n 2. For user-specific queries → use from:@user\n 3. For topic searches → use keywords directly\n 4. For time-based queries → use before:, after:, during:, on:\n 5. For content type → use has:link, has:reaction, etc.\n 6. Combine operators with spaces (implicit AND) or OR\n \n DEFAULTS:\n - limit: 50\n - filename_format: {date}_{channel}_{user}_{id}.txt\n \n Query examples:\n - \"engineering-updates\" → in:#engineering OR in:#engineering-updates\n - \"team-announcements\" → in:#announcements OR in:#general announcement\n - \"from-ceo\" → from:@ceo OR from:@founder\n - \"recent-links\" → has:link after:2026-01-01\n - \"pinned-messages\" → has:pin\n - \"design-feedback\" → in:#design feedback OR in:#design-reviews\n - \"standup-notes\" → standup OR \"daily standup\" OR in:#standup\n - \"customer-issues\" → in:#support OR in:#customer-success issue OR bug\n - \"product-discussions\" → in:#product OR product roadmap OR feature\n - \"onboarding\" → onboarding OR \"new hire\" OR welcome\n - \"decisions\" → decision OR decided OR \"we will\" OR approved\n - \"action-items\" → \"action item\" OR TODO OR \"follow up\" OR assigned\n \n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {channel}, {user}, {date}, {text}\n - For channel messages: {date}_{channel}_{user}_{id}.txt\n - For user messages: {date}_{user}_{channel}_{id}.txt\n - Default: {date}_{channel}_{user}_{id}.txt\n \n Folder/file name: {{ name }}\n {% if guidance %}\n \n Additional guidance from user: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Linear Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferLinearQuery(name: string, guidance: string?) -> LinearQueryResult {\n client AnthropicClient\n prompt #\"\n Convert natural language into a Linear search query.\n \n CORE PRINCIPLE: Text search is powerful - use it liberally. Only add filters when intent is unambiguous.\n \n SYNTAX:\n - Plain text: searches title + description (very flexible, use this a lot!)\n - assignee:name | assignee:me - who it's assigned to\n - no:assignee - unassigned issues\n - state:backlog | state:todo | state:\"in progress\" | state:done | state:canceled\n - priority:1 (urgent) | priority:2 (high) | priority:3 (medium) | priority:4 (low) \n - label:name - filter by label\n - no:label - unlabeled issues\n - team:KEY - filter by team key\n - project:name - filter by project name\n - creator:name - who created it\n - cycle:current | cycle:next | cycle:previous - filter by cycle\n - estimate:N - filter by point estimate\n - OR - combine alternatives (use generously!)\n \n STRATEGY - Think about what the user ACTUALLY wants:\n \n 1. TOPIC/FEATURE searches → mostly text search\n \"checkout bugs\" → checkout bug\n \"api performance\" → api performance\n \"mobile crashes\" → mobile crash\n \"auth issues\" → auth authentication login\n \n 2. PERSON searches → assignee filter + text backup\n \"john's tickets\" → assignee:john OR john\n \"sarah jones work\" → assignee:sarah OR assignee:jones OR sarah OR jones\n \"assigned to mike\" → assignee:mike\n \"unassigned issues\" → no:assignee\n \n 3. STATUS/PRIORITY searches → filters\n \"my open issues\" → assignee:me state:todo OR state:\"in progress\"\n \"urgent bugs\" → priority:1 bug\n \"blocked tickets\" → blocked OR blocking\n \"stuff in review\" → review\n \"in progress issues\" → state:\"in progress\"\n \n 4. PROJECT/CYCLE searches → project or cycle filter\n \"acme project\" → project:acme\n \"current sprint\" → cycle:current\n \"next sprint issues\" → cycle:next\n \"q1 roadmap\" → q1 roadmap\n \"enterprise features\" → enterprise\n \n 5. VAGUE/BROAD searches → text search with synonyms\n \"tech debt\" → \"tech debt\" OR refactor OR cleanup\n \"security\" → security vulnerability\n \"onboarding\" → onboarding OR \"getting started\" OR setup\n \n KEY RULES:\n - When in doubt, use text search - it's flexible\n - Use OR generously to catch variations \n - For multi-word names: assignee:first OR assignee:last (never the full name)\n - Add synonyms for common concepts\n - Don't over-filter - better to return more results than miss things\n \n SEARCH TYPE: \"issues\" (default) or \"projects\" (only if explicitly asking for projects list)\n \n DEFAULTS:\n - limit: 50\n - filename_format: {identifier}_{title}.md\n \n CRITICAL: filename_format MUST contain placeholders like {identifier}, {title}, etc.\n NEVER return a literal filename like \"test.md\" or the folder name as the format.\n The format is a TEMPLATE applied to every result, so it MUST use placeholders.\n \n Folder/file name: {{ name }}\n {% if guidance %}\n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// GitHub Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferGitHubQuery(name: string, guidance: string?) -> GitHubQueryResult {\n client AnthropicClient\n prompt #\"\n Convert natural language into a GitHub query. Think about what an agent would need.\n \n ═══════════════════════════════════════════════════════════════════════════════\n SEARCH TYPES & QUERY FORMATS\n ═══════════════════════════════════════════════════════════════════════════════\n \n 1. REPOS (search_type: \"repos\")\n Listing: list:org/ORGNAME or list:user/USERNAME [type:private|public]\n Search: language:go stars:>100 topic:ml org:company\n \n Examples:\n - \"beam-cloud repos\" → list:org/beam-cloud\n - \"private repos in acme\" → list:org/acme type:private\n - \"popular go projects\" → language:go stars:>1000\n \n 2. ISSUES (search_type: \"issues\") \n Query: repo:owner/name [is:open|closed] [label:X] [assignee:X] [author:X]\n \n Examples:\n - \"open bugs in beta9\" → repo:beam-cloud/beta9 is:open label:bug\n - \"my issues\" → assignee:@me is:open\n - \"security issues in react\" → repo:facebook/react label:security\n \n 3. PULL REQUESTS (search_type: \"prs\")\n Query: repo:owner/name [is:open|closed|merged] [author:X] [review:approved]\n \n Examples:\n - \"open PRs in beta9\" → repo:beam-cloud/beta9 is:open\n - \"merged PRs this week\" → repo:owner/repo is:merged merged:>2026-01-26\n - \"PRs needing review\" → repo:owner/repo is:open review:none\n \n 4. COMMITS (search_type: \"commits\")\n Query: repo:owner/name [branch:X] [author:X] [since:DATE] [path:X]\n \n Examples:\n - \"recent commits in beta9\" → repo:beam-cloud/beta9\n - \"commits to main this week\" → repo:owner/repo branch:main since:2026-01-26\n - \"john's commits\" → repo:owner/repo author:john\n - \"changes to src/api\" → repo:owner/repo path:src/api\n \n 5. RELEASES (search_type: \"releases\")\n Query: repo:owner/name [latest] [prerelease:true|false]\n \n Examples:\n - \"beta9 releases\" → repo:beam-cloud/beta9\n - \"latest release of kubernetes\" → repo:kubernetes/kubernetes latest\n - \"stable releases only\" → repo:owner/repo prerelease:false\n \n 6. WORKFLOW RUNS (search_type: \"workflows\")\n Query: repo:owner/name [status:success|failure|in_progress] [branch:X] [event:push|pr]\n \n Examples:\n - \"failed CI in beta9\" → repo:beam-cloud/beta9 status:failure\n - \"recent builds\" → repo:owner/repo\n - \"PR checks\" → repo:owner/repo event:pull_request\n \n 7. FILES (search_type: \"files\")\n Query: repo:owner/name path:PATH [ref:branch|tag|sha]\n \n Examples:\n - \"beta9 readme\" → repo:beam-cloud/beta9 path:README.md\n - \"package.json in react\" → repo:facebook/react path:package.json\n - \"config files\" → repo:owner/repo path:*.yaml ref:main\n - \"src directory\" → repo:owner/repo path:src/\n \n 8. BRANCHES (search_type: \"branches\")\n Query: repo:owner/name [protected:true|false]\n \n Examples:\n - \"beta9 branches\" → repo:beam-cloud/beta9\n - \"protected branches\" → repo:owner/repo protected:true\n \n ═══════════════════════════════════════════════════════════════════════════════\n CONTENT TYPES (how to format the content)\n ═══════════════════════════════════════════════════════════════════════════════\n \n - \"markdown\" (default) - Rich formatted content, best for agents to read\n - \"diff\" - Code changes (for PRs/commits)\n - \"json\" - Structured metadata\n - \"raw\" - Unprocessed file content (for files)\n \n ═══════════════════════════════════════════════════════════════════════════════\n FILENAME FORMATS (use appropriate placeholders)\n ═══════════════════════════════════════════════════════════════════════════════\n \n Repos: {name}.md or {full_name}.json\n Issues: {number}_{title}.md\n PRs: {number}_{title}.md or {number}_{title}_diff.md\n Commits: {sha_short}_{message}.md\n Releases: {tag}_{name}.md\n Workflows: {id}_{name}_{status}.md\n Files: {path} (preserve original path/name)\n Branches: {name}.md\n \n ═══════════════════════════════════════════════════════════════════════════════\n SMART DETECTION\n ═══════════════════════════════════════════════════════════════════════════════\n \n Detect intent from keywords:\n - \"repos/repositories/projects\" in org/user → repos listing\n - \"prs/pulls/pull requests/merge requests\" → prs\n - \"issues/bugs/tickets/tasks\" → issues\n - \"commits/history/changes/log\" → commits \n - \"releases/versions/tags\" → releases\n - \"ci/cd/builds/workflows/actions/checks\" → workflows\n - \"readme/config/file/code\" → files\n - \"branches\" → branches\n - \"diff/patch/changes\" → content_type: diff\n \n Folder/file name: {{ name }}\n {% if guidance %}\n Additional guidance: {{ guidance }}\n {% endif %}\n \n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Tests\n// -----------------------------------------------------------------------------\n\ntest TestGitHubOpenPRs {\n functions [InferGitHubQuery]\n args {\n name \"open-prs-beta9\"\n guidance null\n }\n}\n\ntest TestGitHubIssues {\n functions [InferGitHubQuery]\n args {\n name \"beta9-bugs\"\n guidance \"show open bugs in the beta9 repo\"\n }\n}\n\ntest TestGitHubRepos {\n functions [InferGitHubQuery]\n args {\n name \"go-cli-tools\"\n guidance null\n }\n}\n\ntest TestGmailUnread {\n functions [InferGmailQuery]\n args {\n name \"unread-emails\"\n guidance null\n }\n}\n\ntest TestGmailFromPerson {\n functions [InferGmailQuery]\n args {\n name \"from-eli\"\n guidance null\n }\n}\n\ntest TestGmailWithGuidance {\n functions [InferGmailQuery]\n args {\n name \"important-emails\"\n guidance \"Only from the last 7 days, max 100 results\"\n }\n}\n\ntest TestGmailInvoices {\n functions [InferGmailQuery]\n args {\n name \"invoices\"\n guidance null\n }\n}\n\ntest TestGmailMeetingNotes {\n functions [InferGmailQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestGmailFlightConfirmation {\n functions [InferGmailQuery]\n args {\n name \"flight-confirmations\"\n guidance null\n }\n}\n\ntest TestGmailReceipts {\n functions [InferGmailQuery]\n args {\n name \"receipts\"\n guidance null\n }\n}\n\ntest TestGmailProjectAlpha {\n functions [InferGmailQuery]\n args {\n name \"project-alpha-emails\"\n guidance null\n }\n}\n\ntest TestGmailShippingUpdates {\n functions [InferGmailQuery]\n args {\n name \"shipping-updates\"\n guidance null\n }\n}\n\ntest TestGmailHealthWithGuidance {\n functions [InferGmailQuery]\n args {\n name \"Medical Documents\"\n guidance \"anything from: summit health, nyu, anything health related basically, mychart\"\n }\n}\n\ntest TestGmailMultiWordSender {\n functions [InferGmailQuery]\n args {\n name \"acme-corp-emails\"\n guidance \"from acme corp and related companies\"\n }\n}\n\ntest TestGmailFuzzySenderNike {\n functions [InferGmailQuery]\n args {\n name \"nike-emails\"\n guidance \"from nike\"\n }\n}\n\ntest TestGmailFuzzySenderAmazon {\n functions [InferGmailQuery]\n args {\n name \"amazon-orders\"\n guidance null\n }\n}\n\ntest TestGDriveShared {\n functions [InferGDriveQuery]\n args {\n name \"shared-with-me\"\n guidance null\n }\n}\n\ntest TestGDrivePdfs {\n functions [InferGDriveQuery]\n args {\n name \"pdfs\"\n guidance null\n }\n}\n\ntest TestGDriveCollegeDocs {\n functions [InferGDriveQuery]\n args {\n name \"college docs\"\n guidance null\n }\n}\n\ntest TestGDriveRecentDocsWithGuidance {\n functions [InferGDriveQuery]\n args {\n name \"recent-docs\"\n guidance \"any docs created in the past week\"\n }\n}\n\ntest TestGDriveOwnedByMe {\n functions [InferGDriveQuery]\n args {\n name \"my docs\"\n guidance null\n }\n}\n\ntest TestNotionMeetings {\n functions [InferNotionQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestEvaluateGmailResults_NoResults {\n functions [EvaluateGmailQueryResults]\n args {\n original_guidance \"anything from summit health, nyu, mychart, health related\"\n generated_query \"from:summit health OR from:nyu OR from:mychart\"\n result_count 0\n sample_results \"[]\"\n }\n}\n\ntest TestEvaluateGmailResults_GoodResults {\n functions [EvaluateGmailQueryResults]\n args {\n original_guidance \"invoices from stripe\"\n generated_query \"invoice from:stripe\"\n result_count 15\n sample_results \"[{\\\"from\\\":\\\"Stripe\\\",\\\"subject\\\":\\\"Your invoice from Acme Corp\\\",\\\"snippet\\\":\\\"Invoice #12345...\\\"},{\\\"from\\\":\\\"Stripe Billing\\\",\\\"subject\\\":\\\"Invoice for January\\\",\\\"snippet\\\":\\\"Amount due: $99\\\"}]\"\n }\n}\n\ntest TestSlackEngineering {\n functions [InferSlackQuery]\n args {\n name \"engineering-updates\"\n guidance null\n }\n}\n\ntest TestSlackFromCEO {\n functions [InferSlackQuery]\n args {\n name \"from-ceo\"\n guidance \"messages from our CEO\"\n }\n}\n\ntest TestSlackRecentLinks {\n functions [InferSlackQuery]\n args {\n name \"recent-links\"\n guidance null\n }\n}\n\ntest TestSlackStandups {\n functions [InferSlackQuery]\n args {\n name \"standup-notes\"\n guidance null\n }\n}\n\ntest TestSlackDecisions {\n functions [InferSlackQuery]\n args {\n name \"team-decisions\"\n guidance \"important decisions made by the team\"\n }\n}\n\ntest TestLinearMyIssues {\n functions [InferLinearQuery]\n args {\n name \"my-issues\"\n guidance null\n }\n}\n\ntest TestLinearMyBugs {\n functions [InferLinearQuery]\n args {\n name \"my-bugs\"\n guidance \"bugs assigned to me\"\n }\n}\n\ntest TestLinearUrgent {\n functions [InferLinearQuery]\n args {\n name \"urgent-issues\"\n guidance null\n }\n}\n\ntest TestLinearTeamBacklog {\n functions [InferLinearQuery]\n args {\n name \"eng-backlog\"\n guidance \"engineering team backlog\"\n }\n}\n\ntest TestLinearInProgress {\n functions [InferLinearQuery]\n args {\n name \"in-progress\"\n guidance null\n }\n}\n\ntest TestPostHogActiveFlags {\n functions [InferPostHogQuery]\n args {\n name \"active-feature-flags\"\n guidance null\n }\n}\n\ntest TestPostHogPageviewEvents {\n functions [InferPostHogQuery]\n args {\n name \"pageview-events\"\n guidance null\n }\n}\n\ntest TestPostHogRetentionInsights {\n functions [InferPostHogQuery]\n args {\n name \"retention-insights\"\n guidance null\n }\n}\n\ntest TestPostHogBetaUsers {\n functions [InferPostHogQuery]\n args {\n name \"beta-users\"\n guidance null\n }\n}\n\ntest TestPostHogErrorEvents {\n functions [InferPostHogQuery]\n args {\n name \"error-events\"\n guidance \"exception and error events from the past week\"\n }\n}\n\ntest TestPostHogSignupFunnel {\n functions [InferPostHogQuery]\n args {\n name \"signup-funnel\"\n guidance null\n }\n}\n\n// -----------------------------------------------------------------------------\n// Confluence Query Inference\n// -----------------------------------------------------------------------------\n\nclass ConfluenceQueryResult {\n cql_query string @description(\"Confluence Query Language (CQL) query string\")\n content_type string @description(\"Content type filter: 'page', 'blogpost', or 'all'. Default: 'all'\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {title}, {space}, {date}, {type}. Example: {space}_{title}_{id}.md\")\n}\n\nfunction InferConfluenceQuery(name: string, guidance: string?) -> ConfluenceQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into a Confluence CQL (Confluence Query Language) query.\n\n CQL SYNTAX:\n - type=page — pages only\n - type=blogpost — blog posts only\n - space=KEY — filter by space key (e.g., space=DEV, space=ENG)\n - title~\"text\" — title contains text (fuzzy match)\n - text~\"text\" — full-text search (searches title + body)\n - label=\"name\" — filter by label\n - creator=\"username\" — filter by creator\n - contributor=\"username\" — filter by contributor\n - lastModified>now(\"-7d\") — modified in the last 7 days\n - lastModified>now(\"-30d\") — modified in the last 30 days\n - lastModified>\"2025-01-01\" — modified after a specific date\n - created>now(\"-7d\") — created in the last 7 days\n - ancestor=12345 — pages under a specific parent\n - AND, OR — combine conditions\n - ORDER BY lastModified DESC, title ASC — sorting\n\n QUERY STRATEGY:\n 1. For topic searches → use text~ for full-text search\n 2. For space-specific → use space=KEY\n 3. For recent content → use lastModified>now(\"-Nd\")\n 4. For labeled content → use label=\"name\"\n 5. For specific content types → use type=page or type=blogpost\n 6. Combine operators with AND/OR\n\n DEFAULTS:\n - content_type: \"all\"\n - limit: 50\n - filename_format: {space}_{title}_{id}.md\n\n Query examples:\n - \"recent-pages\" → lastModified>now(\"-7d\") ORDER BY lastModified DESC, content_type: \"page\"\n - \"engineering-docs\" → (space=ENG OR text~\"engineering\") AND type=page ORDER BY lastModified DESC\n - \"kubernetes-pages\" → text~\"kubernetes\" AND type=page ORDER BY lastModified DESC\n - \"dev-blog-posts\" → space=DEV AND type=blogpost ORDER BY lastModified DESC\n - \"important-docs\" → label=\"important\" ORDER BY lastModified DESC\n - \"api-documentation\" → text~\"api\" AND (label=\"documentation\" OR title~\"api\") ORDER BY lastModified DESC\n - \"onboarding\" → text~\"onboarding\" OR title~\"onboarding\" ORDER BY lastModified DESC\n - \"architecture-decisions\" → text~\"architecture\" OR label=\"adr\" ORDER BY lastModified DESC\n - \"meeting-notes\" → title~\"meeting\" OR label=\"meeting-notes\" ORDER BY lastModified DESC\n - \"recent-blog-posts\" → type=blogpost AND lastModified>now(\"-30d\") ORDER BY lastModified DESC\n - \"project-alpha\" → text~\"project alpha\" ORDER BY lastModified DESC\n\n Filename format: Choose a format that makes sense for the query intent.\n Available placeholders: {id}, {title}, {space}, {date}, {type}\n - For space-specific: {title}_{id}.md\n - For cross-space: {space}_{title}_{id}.md\n - For dated content: {date}_{title}_{id}.md\n - Default: {space}_{title}_{id}.md\n\n Folder/file name: {{ name }}\n {% if guidance %}\n\n Additional guidance from user: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// AgentMail Query Inference\n// -----------------------------------------------------------------------------\n\nclass AgentMailQueryResult {\n agentmail_query string @description(\"Text filter to match against sender, subject, and body. Empty string = all messages (no filter).\")\n inbox_filter string @description(\"Specific inbox email address to query. Empty string = all inboxes.\")\n limit int @description(\"Max results to return (default: 50)\")\n filename_format string @description(\"Template for result filenames. Use {id}, {date}, {from}, {subject}, {inbox}. Example: {date}_{from}_{subject}_{id}.txt\")\n}\n\nfunction InferAgentMailQuery(name: string, guidance: string?) -> AgentMailQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder/file name into an AgentMail query.\n\n AgentMail manages multiple email inboxes (each is an email address like agent@example.com).\n The query is a simple text filter applied client-side against sender, subject, and body.\n There is no server-side search — use simple keywords.\n\n QUERY STRATEGY:\n 1. For general browsing (recent, all, inbox) → empty query, empty inbox_filter\n 2. For sender-specific → use the sender name/email as query text\n 3. For topic-specific → use topic keywords as query text\n 4. For inbox-specific → set inbox_filter to the inbox address if mentioned\n 5. AgentMail has no read/unread distinction — \"unread\" just means recent messages\n\n DEFAULTS:\n - agentmail_query: \"\" (all messages)\n - inbox_filter: \"\" (all inboxes)\n - limit: 50\n - filename_format: {date}_{from}_{subject}_{id}.txt\n\n Query examples:\n - \"recent-emails\" → query: \"\", inbox_filter: \"\"\n - \"unread\" → query: \"\", inbox_filter: \"\"\n - \"emails-from-john\" → query: \"john\", inbox_filter: \"\"\n - \"support-emails\" → query: \"support\", inbox_filter: \"\"\n - \"invoices\" → query: \"invoice\", inbox_filter: \"\"\n - \"meeting-notes\" → query: \"meeting\", inbox_filter: \"\"\n\n Folder/file name: {{ name }}\n {% if guidance %}\n\n Additional guidance from user: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Web Query Inference\n// -----------------------------------------------------------------------------\n\nfunction InferWebQuery(name: string, guidance: string?) -> WebQueryResult {\n client AnthropicClient\n prompt #\"\n Convert a folder name into a web query. Two modes:\n\n web_mode=map — crawl a specific website. web_query must be a valid https:// URL.\n web_mode=search — web search. web_query is a search query string (not a URL).\n\n Use web_mode=map when guidance contains a URL or the name implies a specific site.\n Use web_mode=search when the name describes a topic, question, or information need.\n\n EXAMPLES:\n - \"hennessy-cocktails\" (guidance: \"https://www.hennessy.com/en-us/cocktails\")\n → web_mode: \"map\", web_query: \"https://www.hennessy.com/en-us/cocktails\", include_paths: [\"/en-us/cocktails/*\"], limit: 100\n - \"react-docs\"\n → web_mode: \"map\", web_query: \"https://react.dev/reference\", include_paths: [\"/reference/*\"], limit: 100\n - \"python-tutorial\"\n → web_mode: \"map\", web_query: \"https://docs.python.org/3/tutorial/\", include_paths: [\"/3/tutorial/*\"], limit: 50\n - \"latest AI news\"\n → web_mode: \"search\", web_query: \"latest AI news 2025\", include_paths: [], limit: 20\n - \"best rust web frameworks\"\n → web_mode: \"search\", web_query: \"best rust web frameworks comparison\", include_paths: [], limit: 15\n - \"climate change research papers\"\n → web_mode: \"search\", web_query: \"climate change research papers 2025\", include_paths: [], limit: 20\n\n Defaults: web_mode=map, limit=100, filename_format={title}_{id}.md\n\n Folder name: {{ name }}\n {% if guidance %}\n User guidance: {{ guidance }}\n {% endif %}\n\n {{ ctx.output_format }}\n \"#\n}\n\n// -----------------------------------------------------------------------------\n// Web Query Tests\n// -----------------------------------------------------------------------------\n\ntest TestWebQueryWithGuidance {\n functions [InferWebQuery]\n args {\n name \"hennessy-cocktails\"\n guidance \"https://www.hennessy.com/en-us/cocktails?page=1\"\n }\n}\n\ntest TestWebQueryDocs {\n functions [InferWebQuery]\n args {\n name \"react-docs\"\n guidance null\n }\n}\n\ntest TestWebQueryBlog {\n functions [InferWebQuery]\n args {\n name \"openai-blog\"\n guidance null\n }\n}\n\ntest TestWebQuerySearch {\n functions [InferWebQuery]\n args {\n name \"latest AI news\"\n guidance null\n }\n}\n\ntest TestWebQuerySearchTopic {\n functions [InferWebQuery]\n args {\n name \"best rust web frameworks\"\n guidance null\n }\n}\n\n// -----------------------------------------------------------------------------\n// Confluence Query Tests\n// -----------------------------------------------------------------------------\n\ntest TestConfluenceRecentPages {\n functions [InferConfluenceQuery]\n args {\n name \"recent-pages\"\n guidance null\n }\n}\n\ntest TestConfluenceEngDocs {\n functions [InferConfluenceQuery]\n args {\n name \"engineering-docs\"\n guidance \"documentation from the engineering space\"\n }\n}\n\ntest TestConfluenceKubernetes {\n functions [InferConfluenceQuery]\n args {\n name \"kubernetes-pages\"\n guidance null\n }\n}\n\ntest TestConfluenceMeetingNotes {\n functions [InferConfluenceQuery]\n args {\n name \"meeting-notes\"\n guidance null\n }\n}\n\ntest TestConfluenceArchitectureDecisions {\n functions [InferConfluenceQuery]\n args {\n name \"architecture-decisions\"\n guidance \"ADRs and architecture decision records\"\n }\n}\n\n// -----------------------------------------------------------------------------\n// AgentMail Query Tests\n// -----------------------------------------------------------------------------\n\ntest TestAgentMailRecent {\n functions [InferAgentMailQuery]\n args {\n name \"recent-emails\"\n guidance null\n }\n}\n\ntest TestAgentMailFromJohn {\n functions [InferAgentMailQuery]\n args {\n name \"emails-from-john\"\n guidance null\n }\n}\n\ntest TestAgentMailInvoices {\n functions [InferAgentMailQuery]\n args {\n name \"invoices\"\n guidance null\n }\n}\n\ntest TestAgentMailWithGuidance {\n functions [InferAgentMailQuery]\n args {\n name \"support-emails\"\n guidance \"emails about customer support issues\"\n }\n}\n",
}
</file context>
Comment on lines
+192
to
+198
| inboxes, _, err := a.client.ListInboxes(ctx, 100, "") | ||
| if err != nil { | ||
| return nil, err | ||
| } | ||
| for _, inbox := range inboxes { | ||
| inboxIDs = append(inboxIDs, inbox.InboxID) | ||
| } |
There was a problem hiding this comment.
P2: ExecuteQuery only reads the first page of inboxes, so results silently omit inboxes after the first 100.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At pkg/sources/providers/agentmail.go, line 192:
<comment>ExecuteQuery only reads the first page of inboxes, so results silently omit inboxes after the first 100.</comment>
<file context>
@@ -165,6 +167,165 @@ func (a *AgentMailProvider) Search(_ context.Context, _ *sources.ProviderContext
+ if inboxFilter != "" {
+ inboxIDs = []string{inboxFilter}
+ } else {
+ inboxes, _, err := a.client.ListInboxes(ctx, 100, "")
+ if err != nil {
+ return nil, err
</file context>
Suggested change
| inboxes, _, err := a.client.ListInboxes(ctx, 100, "") | |
| if err != nil { | |
| return nil, err | |
| } | |
| for _, inbox := range inboxes { | |
| inboxIDs = append(inboxIDs, inbox.InboxID) | |
| } | |
| pageToken := "" | |
| for { | |
| inboxes, nextPageToken, err := a.client.ListInboxes(ctx, 100, pageToken) | |
| if err != nil { | |
| return nil, err | |
| } | |
| for _, inbox := range inboxes { | |
| inboxIDs = append(inboxIDs, inbox.InboxID) | |
| } | |
| if nextPageToken == "" { | |
| break | |
| } | |
| pageToken = nextPageToken | |
| } |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary by cubic
Adds AgentMail as a first-class source and tool so you can browse inboxes, read threads and messages, and send/reply via a shared server API key. Also adds cross-provider email thread views (Gmail and Outlook), inbound webhook routing, and runtime routing so replies come from the right inbox.
New Features
agentmailwith a browsable tree:README.md,recent.json,inboxes/{inbox_id}/messages/{sender}/{date_subject}/{meta.json|body.txt}, andinboxes/{inbox_id}/threads/{thread_id}/{meta.json|messages.txt}.ListInboxes,ListMessages,GetMessage,ListThreads,GetThread,ReplyToMessage.to/reply_to, and forwards to agents.AIRSTORE_AGENT_ROUTING_JSON, injects a prompt hint, and binds a default sender inbox from agent/workspace channel bindings.agentmailfilter (inbox/from/subject).agentmailquery inference and default filename format support.list-messages,get-message,get-thread,send,reply. Gateway builds summaries forsend/reply. Tools run with server-level auth (AuthNone) and don’t require user connections. Auto-registersagentmailsource and tools when configured.Migration
channels.agentMail.apiKey,channels.agentMail.baseURL, andchannels.agentMail.domainin server config to enable. No user OAuth; uses shared server credentials.Written for commit 59586a8. Summary will update on new commits.