release: v0.7.3

anandgupta42 · claude · anandgupta42 · commit 56259a33985d · 2026-05-23T21:07:32.000-07:00
Telemetry-driven hardening release. Five P0 fixes from the telemetry-analysis-2026-05-21 pass (#828 finops auto-pick, #831 project_scan defensive spawn, #833 build-agent name normalization, #837 tokens_input_total semantics, #839 webfetch 404 cache) plus a multi-persona pre-release review that surfaced privacy/security hardening: credential strip on git.remoteUrl, masked git stderr, agent-name input hardening (control-char strip + NFKC + length cap), auth-bearing query-param strip from the webfetch cache key, cache-hit error prefix, warehouse_filter param description disambiguation, DEFAULT_FINOPS_TYPES single source of truth, and docs alignment for the auto-pick behavior. 34 adversarial tests pin every invariant. 4 deferred items filed as #840 #841 #842 #843. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
diff --git a/.gitignore b/.gitignore
@@ -44,3 +44,4 @@ __pycache__
 **/.github/meta/commit.txt
 **/.github/meta/diff.txt
 **/.github/meta/pr-body-*.md
+*.bak
diff --git a/CHANGELOG.md b/CHANGELOG.md
diff --git a/docs/docs/data-engineering/tools/finops-tools.md b/docs/docs/data-engineering/tools/finops-tools.md
@@ -2,6 +2,9 @@
 
 Cost optimization and warehouse governance. These tools help you find where money is being wasted and fix it.
 
+!!! note "v0.7.3+ — warehouse parameter is now optional"
+    Every `finops_*` tool used to require an explicit `warehouse=<connection-name>` argument. As of v0.7.3, the parameter is optional — if omitted, the tool auto-picks the first configured warehouse whose driver type supports the operation. Pass `warehouse=` explicitly when you have multiple compatible connections and want a specific one. The bare form (`finops_query_history --days 7`) works.
+
 ## finops_query_history
 
 Fetch recent query execution history from your warehouse.
@@ -27,11 +30,11 @@ Summary:
 ```
 
 **Parameters:**
-- `warehouse` (required): Connection name
+- `warehouse` (optional, since v0.7.3): Connection name. If omitted, the first configured Snowflake / BigQuery / Databricks / PostgreSQL warehouse is auto-picked.
 - `days` (optional, default: 7): Lookback period
 - `limit` (optional, default: 100): Max queries returned
 - `user` (optional): Filter by username
-- `warehouse_filter` (optional): Filter by compute warehouse name
+- `warehouse_filter` (optional, Snowflake-only): Filter the result rows by the in-warehouse compute name (different from the `warehouse` connection parameter). Use `warehouse` to pick which connection to query; use `warehouse_filter` to narrow which Snowflake virtual warehouse's queries appear in the output.
 
 **Data sources by warehouse:**
 - Snowflake: `QUERY_HISTORY` function
@@ -46,7 +49,8 @@ Summary:
 Break down credit consumption by warehouse, time, and user.
 
 ```
-> finops_analyze_credits prod-snowflake --days 30
+> finops_analyze_credits --days 30
+# equivalent: finops_analyze_credits prod-snowflake --days 30
 
 Credit Analysis (last 30 days):
 
diff --git a/packages/opencode/src/altimate/native/finops/credit-analyzer.ts b/packages/opencode/src/altimate/native/finops/credit-analyzer.ts
@@ -6,15 +6,15 @@
 
 import * as Registry from "../connections/registry"
 import { bqRegionFor, interpolateBqRegion } from "./bq-utils"
-import { resolveFinopsWarehouse } from "./warehouse-resolver"
+import { resolveFinopsWarehouse, DEFAULT_FINOPS_TYPES } from "./warehouse-resolver"
 import type {
   CreditAnalysisParams,
   CreditAnalysisResult,
   ExpensiveQueriesParams,
   ExpensiveQueriesResult,
 } from "../types"
 
-const CREDIT_SUPPORTED_TYPES = ["snowflake", "bigquery", "databricks"] as const
+const CREDIT_SUPPORTED_TYPES = DEFAULT_FINOPS_TYPES
 
 // ---------------------------------------------------------------------------
 // Snowflake SQL templates
diff --git a/packages/opencode/src/altimate/native/finops/role-access.ts b/packages/opencode/src/altimate/native/finops/role-access.ts
@@ -6,7 +6,7 @@
 
 import * as Registry from "../connections/registry"
 import { bqRegionFor, interpolateBqRegion } from "./bq-utils"
-import { resolveFinopsWarehouse } from "./warehouse-resolver"
+import { resolveFinopsWarehouse, DEFAULT_FINOPS_TYPES } from "./warehouse-resolver"
 import type {
   RoleGrantsParams,
   RoleGrantsResult,
@@ -108,7 +108,7 @@ LIMIT ?
 // Helpers
 // ---------------------------------------------------------------------------
 
-const GRANTS_SUPPORTED_TYPES = ["snowflake", "bigquery", "databricks"] as const
+const GRANTS_SUPPORTED_TYPES = DEFAULT_FINOPS_TYPES
 const SNOWFLAKE_ONLY_TYPES = ["snowflake"] as const
 
 function rowsToRecords(result: { columns: string[]; rows: any[][] }): Record<string, unknown>[] {
diff --git a/packages/opencode/src/altimate/native/finops/unused-resources.ts b/packages/opencode/src/altimate/native/finops/unused-resources.ts
@@ -6,13 +6,13 @@
 
 import * as Registry from "../connections/registry"
 import { bqRegionFor, interpolateBqRegion } from "./bq-utils"
-import { resolveFinopsWarehouse } from "./warehouse-resolver"
+import { resolveFinopsWarehouse, DEFAULT_FINOPS_TYPES } from "./warehouse-resolver"
 import type {
   UnusedResourcesParams,
   UnusedResourcesResult,
 } from "../types"
 
-const UNUSED_RESOURCES_SUPPORTED_TYPES = ["snowflake", "bigquery", "databricks"] as const
+const UNUSED_RESOURCES_SUPPORTED_TYPES = DEFAULT_FINOPS_TYPES
 
 // ---------------------------------------------------------------------------
 // Snowflake SQL templates
diff --git a/packages/opencode/src/altimate/native/finops/warehouse-advisor.ts b/packages/opencode/src/altimate/native/finops/warehouse-advisor.ts
@@ -6,7 +6,7 @@
 
 import * as Registry from "../connections/registry"
 import { bqRegionFor, interpolateBqRegion } from "./bq-utils"
-import { resolveFinopsWarehouse } from "./warehouse-resolver"
+import { resolveFinopsWarehouse, DEFAULT_FINOPS_TYPES } from "./warehouse-resolver"
 import type {
   WarehouseAdvisorParams,
   WarehouseAdvisorResult,
@@ -123,7 +123,7 @@ const SIZE_ORDER = ["X-Small", "Small", "Medium", "Large", "X-Large", "2X-Large"
 // Helpers
 // ---------------------------------------------------------------------------
 
-const ADVISOR_SUPPORTED_TYPES = ["snowflake", "bigquery", "databricks"] as const
+const ADVISOR_SUPPORTED_TYPES = DEFAULT_FINOPS_TYPES
 
 function buildLoadSql(whType: string, days: number, bqRegion?: unknown): string | null {
   if (whType === "snowflake") return SNOWFLAKE_LOAD_SQL.replace("{days}", String(days))
diff --git a/packages/opencode/src/altimate/native/finops/warehouse-resolver.ts b/packages/opencode/src/altimate/native/finops/warehouse-resolver.ts
@@ -18,6 +18,17 @@
 
 import * as Registry from "../connections/registry"
 
+/**
+ * Default supported driver types for finops operations that work against any
+ * cloud warehouse with usage/billing tables (credit analysis, sizing advice,
+ * unused-resource detection, etc.). Centralized here so adding a new
+ * warehouse driver doesn't require touching 3+ handler files.
+ *
+ * Operations with narrower support (e.g. Snowflake-only role hierarchy,
+ * postgres-aware query history) declare their own const arrays.
+ */
+export const DEFAULT_FINOPS_TYPES = ["snowflake", "bigquery", "databricks"] as const
+
 export type FinopsWarehouseResolution =
   | { kind: "ok"; warehouse: string; type: string; autoPicked: boolean }
   | { kind: "error"; error: string }
diff --git a/packages/opencode/src/altimate/tools/finops-analyze-credits.ts b/packages/opencode/src/altimate/tools/finops-analyze-credits.ts
@@ -73,7 +73,14 @@ export const FinopsAnalyzeCreditsTool = Tool.define("finops_analyze_credits", {
       ),
     days: z.number().optional().default(30).describe("Days of history to analyze"),
     limit: z.number().optional().default(50).describe("Max daily records"),
-    warehouse_filter: z.string().optional().describe("Filter to a specific Snowflake warehouse"),
+    warehouse_filter: z
+      .string()
+      .optional()
+      .describe(
+        "Filter result rows by an in-warehouse Snowflake compute name (NOT the connection name). " +
+          "Distinct from `warehouse`: `warehouse` picks which connection to query; `warehouse_filter` narrows " +
+          "which Snowflake virtual warehouse's queries are returned in the result set. Snowflake only.",
+      ),
   }),
   async execute(args, ctx) {
     try {
diff --git a/packages/opencode/src/altimate/tools/finops-query-history.ts b/packages/opencode/src/altimate/tools/finops-query-history.ts
@@ -61,7 +61,14 @@ export const FinopsQueryHistoryTool = Tool.define("finops_query_history", {
     days: z.number().optional().default(7).describe("How many days of history to fetch"),
     limit: z.number().optional().default(100).describe("Maximum number of queries to return"),
     user: z.string().optional().describe("Filter to a specific user (Snowflake only)"),
-    warehouse_filter: z.string().optional().describe("Filter to a specific warehouse name (Snowflake only)"),
+    warehouse_filter: z
+      .string()
+      .optional()
+      .describe(
+        "Filter result rows by an in-warehouse Snowflake compute name (NOT the connection name). " +
+          "Distinct from `warehouse`: `warehouse` picks which connection to query; `warehouse_filter` narrows " +
+          "which Snowflake virtual warehouse's queries are returned in the result set. Snowflake only.",
+      ),
   }),
   async execute(args, ctx) {
     try {
diff --git a/packages/opencode/src/altimate/tools/project-scan.ts b/packages/opencode/src/altimate/tools/project-scan.ts
@@ -132,10 +132,18 @@ export async function detectGit(): Promise<GitInfo> {
     if (isRepoResult.exitCode === 128) {
       return { isRepo: false, gitAvailable: true }
     }
+    // Run stderr through the telemetry masker BEFORE persisting so emails,
+    // bearer tokens, and api-keys can't leak into LLM-visible metadata,
+    // session transcripts, or App Insights via `metadata.git.gitError`.
+    // Cap at 240 chars after masking (mask replacements can shorten or
+    // lengthen the input minimally).
+    const maskedStderr = isRepoResult.stderr
+      ? Telemetry.maskString(isRepoResult.stderr.trim()).slice(0, 240) || undefined
+      : undefined
     return {
       isRepo: false,
       gitAvailable: true,
-      gitError: { exitCode: isRepoResult.exitCode, stderr: isRepoResult.stderr?.trim() || undefined },
+      gitError: { exitCode: isRepoResult.exitCode, stderr: maskedStderr },
     }
   }
 
@@ -146,12 +154,40 @@ export async function detectGit(): Promise<GitInfo> {
   let remoteUrl: string | undefined
   const remoteResult = safeSpawnSync(["git", "remote", "get-url", "origin"])
   if (remoteResult && remoteResult.exitCode === 0) {
-    remoteUrl = remoteResult.stdout.trim()
+    remoteUrl = stripGitRemoteCredentials(remoteResult.stdout.trim())
   }
 
   return { isRepo: true, gitAvailable: true, branch, remoteUrl }
 }
 
+/**
+ * Strip embedded credentials from a git remote URL before surfacing it.
+ *
+ * `git remote get-url origin` returns the URL verbatim including HTTPS
+ * basic-auth (e.g. `https://x-access-token:ghp_xxx@github.com/...`) when
+ * the user committed creds into the remote. `remoteUrl` flows into tool
+ * result metadata that ships to the LLM provider AND is persisted in
+ * session transcripts — leaving creds in there is a recurring exfil path.
+ *
+ * SSH-form remotes (`git@github.com:owner/repo.git`) have no userinfo
+ * concept and are left untouched. URLs we can't parse are dropped to
+ * undefined (better to lose the breadcrumb than leak creds).
+ */
+function stripGitRemoteCredentials(url: string): string | undefined {
+  if (!url) return undefined
+  // SSH form: `git@host:path` — no creds to strip.
+  if (/^[\w.-]+@[\w.-]+:/.test(url) && !url.includes("://")) return url
+  try {
+    const parsed = new URL(url)
+    parsed.username = ""
+    parsed.password = ""
+    return parsed.toString()
+  } catch {
+    // Unparseable — drop entirely rather than risk leaking embedded creds.
+    return undefined
+  }
+}
+
 export async function detectDbtProject(startDir: string): Promise<DbtProjectInfo> {
   let dir = startDir
   for (let i = 0; i < 5; i++) {
diff --git a/packages/opencode/src/session/prompt.ts b/packages/opencode/src/session/prompt.ts
@@ -84,12 +84,25 @@ export namespace SessionPrompt {
   // emits below so they can never drift. Future telemetry events with an `agent`
   // field should route through this helper too.
   function normalizeAgentName(name: string | undefined): string {
-    // Case-insensitive guard: a future config, custom prompt, or hand-edited
-    // persisted session could surface "Build" or "BUILD" and the phantom
-    // bucket would come back. One toLowerCase() costs nothing and pins the
-    // contract regardless of caller hygiene.
-    if (!name || name.toLowerCase() === "build") return "builder"
-    return name
+    // Defence-in-depth before the legacy-name compare:
+    //   1. Strip C0 control characters (\x00-\x1f) — neutralizes log-injection
+    //      via embedded newlines/CRs that would split the telemetry field into
+    //      two fake events on App Insights.
+    //   2. Unicode-normalize (NFKC) — collapses visually-identical homoglyphs
+    //      so "ｂｕｉｌｄｅｒ" (fullwidth) doesn't create a separate bucket.
+    //   3. Cap at 64 chars — agent names should be slugs; anything larger is
+    //      a cardinality bomb or an injection vector. The agent registry's
+    //      longest legitimate name is well under this cap.
+    if (!name) return "builder"
+    const cleaned = name
+      .replace(/[\x00-\x1f\x7f]/g, "")
+      .normalize("NFKC")
+      .slice(0, 64)
+    // Case-insensitive legacy-name guard: a future config, custom prompt, or
+    // hand-edited persisted session could surface "Build"/"BUILD" and the
+    // phantom telemetry bucket would come back.
+    if (!cleaned || cleaned.toLowerCase() === "build") return "builder"
+    return cleaned
   }
   // altimate_change end
 
diff --git a/packages/opencode/src/tool/webfetch.ts b/packages/opencode/src/tool/webfetch.ts
@@ -101,6 +101,35 @@ const TRACKING_PARAMS = new Set([
   "piwik_keyword",
 ])
 
+// Auth-bearing query params. These are *functional* (they change what the
+// server returns — a different signature accesses a different resource view)
+// so they're NOT in TRACKING_PARAMS (which is "safe to drop, doesn't change
+// the resource"). But for cache-key purposes we DO want to drop them:
+//   1. Compliance — presigned-S3 URLs, GCS signed URLs, and bearer-token
+//      query strings shouldn't sit in an in-memory Map for 30 min where
+//      heap dumps, debug bundles, and process snapshots could capture them.
+//   2. Equivalence — a 404 on `?signature=A` strongly implies a 404 on
+//      `?signature=B` for the same path (the path is wrong, not the sig),
+//      so collapsing them is the *right* cache behaviour.
+// We do NOT strip these from the URL we actually fetch — only from the
+// cache key. That happens via `AUTH_PARAMS_FOR_CACHE_KEY` in
+// `normalizeUrlForCache` below.
+const AUTH_PARAMS_FOR_CACHE_KEY = new Set([
+  "token",
+  "access_token",
+  "id_token",
+  "api_key",
+  "apikey",
+  "key", // GCS / GCP signed URLs
+  "signature",
+  "sig",
+  "x-amz-signature",
+  "x-amz-credential",
+  "x-amz-security-token",
+  "x-goog-signature",
+  "x-goog-credential",
+])
+
 function isTrackingParamPrefix(name: string): boolean {
   return name.startsWith("__cf_chl_") || name.startsWith("utm_")
 }
@@ -123,6 +152,10 @@ export function normalizeUrlForCache(url: string): string {
     const kept: [string, string][] = []
     for (const [k, v] of parsed.searchParams) {
       if (TRACKING_PARAMS.has(k) || isTrackingParamPrefix(k)) continue
+      // Drop auth-bearing params from the cache key — see
+      // AUTH_PARAMS_FOR_CACHE_KEY comment for the why. The actual fetch
+      // still uses the raw URL with creds intact.
+      if (AUTH_PARAMS_FOR_CACHE_KEY.has(k.toLowerCase())) continue
       kept.push([k, v])
     }
     // Sort by (key, value) so duplicate-key params produce a stable order:
@@ -146,15 +179,16 @@ function ttlForStatus(status: number): number {
   return status === 451 ? FAILURE_CACHE_TTL_451 : FAILURE_CACHE_TTL
 }
 
-function isUrlCachedFailure(url: string): { status: number } | null {
+function isUrlCachedFailure(url: string): { status: number; ageMs: number } | null {
   const key = normalizeUrlForCache(url)
   const entry = failedUrls.get(key)
   if (!entry) return null
-  if (Date.now() - entry.timestamp > ttlForStatus(entry.status)) {
+  const ageMs = Date.now() - entry.timestamp
+  if (ageMs > ttlForStatus(entry.status)) {
     failedUrls.delete(key)
     return null
   }
-  return { status: entry.status }
+  return { status: entry.status, ageMs }
 }
 
 const MAX_CACHED_URLS = 500
@@ -222,7 +256,7 @@ export function _failureCacheKeysInOrder(): string[] {
  * Probe the cache for a URL — same code path as production lookups.
  * @internal — only used by tests; do not call from production code.
  */
-export function _isUrlCachedFailureForTest(url: string): { status: number } | null {
+export function _isUrlCachedFailureForTest(url: string): { status: number; ageMs: number } | null {
   return isUrlCachedFailure(url)
 }
 
@@ -280,10 +314,15 @@ export const WebFetchTool = Tool.define("webfetch", {
       throw new Error(`Invalid URL: "${params.url.slice(0, 200)}" is not a valid URL. Check the format and try again.`)
     }
 
-    // Check failure cache — avoid re-fetching URLs that already returned 404/410
+    // Check failure cache — avoid re-fetching URLs that already returned 404/410/451.
+    // Prefix with "(cached failure, Nm ago)" so the agent can tell a cache-hit
+    // refusal apart from a live network call. Otherwise an LLM debugging
+    // "why is this instant?" has no signal that the URL was normalized and
+    // collapsed onto a prior failure.
     const cached = isUrlCachedFailure(params.url)
     if (cached) {
-      throw new Error(buildFetchError(params.url, cached.status))
+      const ageMin = Math.max(1, Math.round(cached.ageMs / 60_000))
+      throw new Error(`(cached failure, ${ageMin}m ago) ${buildFetchError(params.url, cached.status)}`)
     }
     // altimate_change end
 
diff --git a/packages/opencode/test/session/plan-layer-e2e.test.ts b/packages/opencode/test/session/plan-layer-e2e.test.ts
@@ -327,11 +327,32 @@ describe("sessionAgentName fix safety", () => {
     )
     const declIdx = promptTs.indexOf("function normalizeAgentName")
     expect(declIdx).toBeGreaterThan(-1)
-    const body = promptTs.slice(declIdx, declIdx + 400)
+    const body = promptTs.slice(declIdx, declIdx + 1200)
     // The body should case-fold before comparing — either toLowerCase() or
-    // toUpperCase() before the equality check. Anchored so a refactor to a
-    // raw string compare fails this test.
-    expect(body).toMatch(/name\.toLowerCase\(\)\s*===\s*"build"/)
+    // toUpperCase() before the equality check. Anchored to the legacy-name
+    // check specifically (variable name allowed to drift, the .toLowerCase()
+    // adjacent to "build" is what matters).
+    expect(body).toMatch(/\.toLowerCase\(\)\s*===\s*"build"/)
+  })
+
+  test("normalizeAgentName hardens against adversarial input (control chars, length, unicode)", async () => {
+    // Reviewer (compliance / chaos, 2026-05-24) flagged that the helper
+    // accepted arbitrary user-supplied agent names without sanitization:
+    // a 50KB string, embedded newlines (log-injection on App Insights),
+    // or fullwidth homoglyphs would create separate telemetry buckets
+    // or split events. Pin that the helper:
+    //   - strips control chars
+    //   - NFKC-normalizes
+    //   - caps length
+    const promptTs = await fs.readFile(
+      path.join(__dirname, "../../src/session/prompt.ts"),
+      "utf-8",
+    )
+    const declIdx = promptTs.indexOf("function normalizeAgentName")
+    const body = promptTs.slice(declIdx, declIdx + 1200)
+    expect(body).toMatch(/\[\\x00-\\x1f/) // C0 strip
+    expect(body).toMatch(/normalize\(\s*["']NFKC["']\s*\)/)
+    expect(body).toMatch(/\.slice\(\s*0\s*,\s*\d{2,3}\s*\)/) // length cap (2-3 digit, e.g. 64)
   })
 })
 
diff --git a/packages/opencode/test/skill/release-v0.7.3-adversarial.test.ts b/packages/opencode/test/skill/release-v0.7.3-adversarial.test.ts
diff --git a/packages/opencode/test/tool/webfetch-cache.test.ts b/packages/opencode/test/tool/webfetch-cache.test.ts
