feat(showcase): live provider:'mcp' declarative connector demo + CI-pinned dogfood proof (#3056)#3062
Merged
Merged
Conversation
…e upstream (#3017) ADR-0097 made every declarative-connector materialization failure fatal at boot — right for configuration faults, wrong for operational ones: a provider: 'mcp' instance must contact its MCP server (tools/list) to materialize, so a transient network blip aborted the whole app boot. - spec: ConnectorUpstreamUnavailableError (code CONNECTOR_UPSTREAM_UNAVAILABLE, structural guard isConnectorUpstreamUnavailable) lets a provider factory mark a failure as 'upstream temporarily unreachable — degrade and retry' instead of fatal. New integration/connector-provider-errors.ts; api-surface +3. - service-automation: the reconcile degrades such instances in BOTH modes. With no live connector under the name it registers an action-less husk — state: 'degraded' + degradedReason on the GET /connectors descriptor, status: 'error' on the def — so the instance stays visible instead of silently missing; connector_action dispatch fails with the reason and a 'retries automatically' pointer. On a changed-config re-materialization the old connector keeps serving. Degraded instances retry on an exponential backoff (5s doubling to 5min, reset by config edits) and on every metadata:reloaded reconcile; recovery swaps the husk atomically. Reconcile runs (boot / reload / retry timer) are serialized; destroy() cancels the retry loop. - connector-mcp: connect / tools/list failures are classified unavailable; transport-shape validation stays a plain (fatal) throw. Configuration faults (unknown provider, invalid providerConfig, unresolvable credentialRef, name conflicts) keep the ADR-0097 fail-loud contract, verified by the existing tests. ADR-0097 gains an 'Upstream availability' section; the deferred live-mcp showcase demo and the openapi remote-URL classification are recorded as scope boundaries. Tests: 10 new reconcile cases (degrade visibility, pointed dispatch error, fake-timer recovery, exponential backoff, reload retry, removal while degraded, changed-config keeps old serving, revert cancels retry, shutdown cancels retry, plain-throw still fatal); 3 new mcp-provider classification cases; 3 new spec error-contract cases. Refs #3017 (mechanism; showcase demo deferred — see ADR scope boundaries) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pbmw3pMqNJfPbkvwh9Fkjs
…lt-deny + host allowlist (#3055) A declarative provider: 'mcp' entry with a stdio transport spawns a local child process from METADATA — which a runtime Studio publish can introduce (the reload reconcile materializes new instances; soft mode skips failures, not successes). Anyone who can publish metadata could otherwise execute arbitrary commands on the server. Declarative stdio is now denied by default and opt-in per host: new ConnectorMcpPlugin({ declarativeStdio: ['my-mcp-server'] }) // allowlist new ConnectorMcpPlugin({ declarativeStdio: true }) // full trust - createMcpProviderFactory gains a declarativeStdio policy (undefined/false = deny, string[] = strict-equality command allowlist, true = allow any); ConnectorMcpPlugin plumbs it through. New export McpDeclarativeStdioPolicy. - A violation is a CONFIGURATION fault: plain throw with an actionable opt-in message (fatal at boot, skipped+logged on reload) — never classified CONNECTOR_UPSTREAM_UNAVAILABLE, so a security rejection cannot be retried into existence (#3049 seam). - http transports are unaffected (same trust class as the http node), and so are hand-wired connectors (plugin instance options / createMcpConnector) — their command lives in host code, a different trust anchor than metadata. - ADR-0097 gains a 'Declarative stdio policy' section (cross-ref ADR-0024 §4, honest about the allowlist being a coarse boundary); this gate is the precondition for shipping connector-mcp in default presets (#3056). Tests: default-deny (plain, non-retryable, rejected before any connection attempt), allowlist hit/miss, true passthrough, http unaffected, plugin option plumb-through, hand-wired stdio path explicitly un-gated (e2e). Refs #3055. Stacked on #3049 (claude/adr-0097-mcp-nonfatal-boot). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pbmw3pMqNJfPbkvwh9Fkjs
#3049's squash (4f8c2d1) is content-identical to this branch's base commit, so this merge only rebases the PR surface — the three-dot diff vs main now shows the #3055 gate alone. # Conflicts: # docs/adr/0097-declarative-connector-instances.md # packages/connectors/connector-mcp/src/mcp-provider.test.ts # packages/connectors/connector-mcp/src/mcp-provider.ts
…ve provider:'mcp' showcase demo (#3056) Closes the last authoring gap in ADR-0097's 'expressible AND executable as pure metadata' promise, and lands the provider:'mcp' acceptance demo the ADR originally envisioned (deferred at #3017). create-objectstack (blank template): - plugins: gains new ConnectorRestPlugin() / ConnectorOpenApiPlugin() / ConnectorMcpPlugin() (zero-arg = provider factory only), so a scaffolded app materializes declarative provider-bound connectors with no plugin-wiring step. README documents the default providers, credentialRef rule, and the #3055 declarativeStdio opt-in. Deps added at ^15.0.0. examples/app-showcase (live mcp demo): - scripts/mcp-fixture.mjs — tiny in-repo stdio MCP server (one deterministic echo_upper tool): no network, no ports, no boot-ordering coupling, so the demo is CI-deterministic. - DevToolsMcpConnector (provider: 'mcp', stdio transport 'node ./scripts/mcp-fixture.mjs') + ShowcaseMcpConnectorEchoFlow dispatching echo_upper end-to-end; coverage notes updated. - objectstack.config.ts wires new ConnectorMcpPlugin({ declarativeStdio: ['node'] }) — dogfooding the #3055 default-deny opt-in, with the coarse-boundary caveat spelled out. dogfood: - showcase-declarative-mcp.dogfood.test.ts boots the REAL showcase (chdir to the app root, exactly how os dev runs), injects the three executors, and pins: GET /connectors lists showcase_mcp_tools origin=declarative state=ready with echo_upper (rest + openapi siblings ready too; catalog descriptor absent), and the flow trigger round-trips OBJECTSTACK from the fixture's tools/call into the run output. docs: flows.mdx explains where connector_action connectors come from, the default preset, and the stdio opt-in. Verified: dogfood proof 2/2; turbo test dogfood+showcase+create-objectstack 63 tasks green; showcase validate (23 flows / 8 plugins) + typecheck clean; eslint clean; lockfile updated for the new deps. Refs #3056 (template + showcase + demo + docs; issue also tracks nothing further after this — see closing comment) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pbmw3pMqNJfPbkvwh9Fkjs
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
github-actions
Bot
added
size/m
documentation
Contributor
📓 Docs Drift CheckThis PR changes 1 package(s): 2 hand-written doc(s) reference the affected code and may need an implementation-accuracy re-verification:
|
…he 15.1.0 publish The scaffold-e2e job installs the generated app from the npm REGISTRY, and ConnectorOpenApiPlugin (and the #3055 declarativeStdio option) only exist in the unpublished 15.1.0 (queued in the Changesets release PR). A template must not reference unpublished APIs — the same failure would hit real 'npm create objectstack' users. Reverted the template files + changeset; once 15.1.0 publishes, the template's ^15.0.0 pins resolve to it and the preset piece lands as a small follow-up. flows.mdx reworded to describe the mechanism without the scaffold claim. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Pbmw3pMqNJfPbkvwh9Fkjs
os-zhuang
changed the title
os-zhuang
marked this pull request as ready for review
July 16, 2026 14:09
os-zhuang
pushed a commit
that referenced
this pull request
Jul 16, 2026
…dogfood test + regen lockfile after merge showcase-declarative-mcp.dogfood.test.ts landed on main (#3062) against the old packages/dogfood location; after the packages/qa migration its '../../../examples/app-showcase' URL resolved inside packages/. One more level, same as the other conformance tests. Lockfile regenerated after taking main's side in the merge. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu
os-zhuang
added a commit
that referenced
this pull request
Jul 16, 2026
…结论 (#2462) (#3037) * feat(transport): prove the IHttpServer port is Hono-free — second adapter on raw node:http (ADR-0076 D11/OQ#10, #2462) Multi-adapter was designed but unproven: only the Hono adapter existed, and ADR-0076 D11 flagged 'the normalized context shows Hono-isms' as a caveat blocking the transport decomposition. This adds the validation: - New @objectstack/plugin-node-server: a thin IHttpServer implementation on raw node:http with ZERO dependencies beyond @objectstack/core — :param + trailing-* routing (registration-order first-match, same as the primary adapter), eager JSON/urlencoded body parse with lazy rawBody() for binary, SSE via native res.write/end, 404/405-with-Allow semantics, EADDRINUSE retry, graceful drain on close. Deliberately no getRawApp()/mount(). - Cross-adapter conformance suite: boots the dispatcher bridge AND the REST route generator (with ObjectQL + memory driver) on BOTH adapters over real sockets — /data CRUD roundtrip, /meta reads, /ready, /health, discovery, :param routing, 404/405 parity, plus a probe-for-probe response-shape parity matrix between the two adapters. 40 assertions, all green. Findings recorded in ADR-0076 (OQ#9/OQ#10/OQ#11 resolution notes): - The port has NO hard Hono-isms. The Host-header backfill in the Hono adapter is a Fetch-API artifact local to that adapter, not a port leak. - All remaining Hono coupling is confined to the feature-detected getRawApp() escape hatch (metadata HMR, cloud-connection routes, static/SPA + CORS + Server-Timing) which degrades gracefully. - Follow-up for the D11 window: codify res.write/res.end (SSE) and getPort() — extensions consumers already rely on — plus 404/405 semantics into the IHttpServer contract. - OQ#9 audit verdict (delineate, don't merge; only discovery + packages are genuinely double-mounted) recorded in the ADR open-questions list. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(plugin-node-server): harden query/params against prototype pollution (CodeQL) CodeQL flagged remote property injection: query-param keys come straight from the request URL, so building the map on a plain object literal let '?__proto__=…' write through the prototype chain. Use null-prototype objects for query and params and drop the dangerous keys (__proto__/constructor/prototype) outright; regression test pins that '?__proto__=polluted' is dropped and Object.prototype stays clean. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(plugin-node-server): eliminate the dynamic property-write sink CodeQL flags The blacklist guard added in the previous commit was not recognized as a sanitizer by CodeQL's js/remote-property-injection rule — the sink is the computed write itself (obj[userKey] = …). Build query and params via Object.fromEntries (own data properties only, no prototype-chain walk) so the sink no longer exists; the __proto__/constructor/prototype drop stays as defense in depth. Regression test unchanged and green (41/41). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * refactor(qa): reposition the port-conformance adapter as a private QA gate — packages/qa/http-conformance Maintainer review: the node:http adapter is a validation instrument, not a product server, so it must not ship as a publishable plugin- package. Reposition it accordingly: - Move packages/plugins/plugin-node-server → packages/qa/http-conformance; rename @objectstack/plugin-node-server → @objectstack/http-conformance, private: true, test-only (no build/publish), matching the existing verification-gate packages (dogfood, downstream-contract). - packages/qa/ becomes the unified home for non-published verification gates (new pnpm-workspace glob); migrating dogfood/downstream-contract there is proposed as a follow-up. - Drop the changeset + fixed-group entry (private packages are not versioned); update the ADR-0076 OQ#10 note to the new name. Behavior unchanged: same adapter, same 41-assertion cross-adapter conformance suite, all green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * refactor(qa): migrate dogfood + downstream-contract into packages/qa/ (unified verification-gate home) Completes the packages/qa/ consolidation started with http-conformance: all three non-published verification gates (dogfood regression gate, downstream-contract compatibility gate, http-conformance port gate) now live under one directory. Mechanical path migration — no behavior change: - git mv packages/{dogfood,downstream-contract} → packages/qa/ - rewrite every 'packages/dogfood' / 'packages/downstream-contract' breadcrumb: spec liveness proof registry + JSON ledgers, spec-liveness-check.yml path trigger, eslint ignore, docs/ADRs, source comments - bump the repo-root traversal one level in the four dogfood conformance tests that resolve REPO_ROOT from __dirname Verified from the new locations: dogfood 273 passed / 3 skipped, downstream-contract typecheck + 14 tests, spec check:liveness resolves all bound proofs, http-conformance 41/41. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * chore: retrigger CI — Actions did not fire on dc6507c Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * docs(changeset): fix dogfood path breadcrumb after packages/qa migration Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(ci): build the examples' dependency closure explicitly in the typecheck job + correct stale ADR-0076 D12 status Two follow-ups after merging latest main: - lint.yml: the example-app typecheck step built only ./packages/* (direct children). The connector packages the showcase imports were built purely by accident — through dogfood's dependency chain — which broke when dogfood moved to packages/qa/. Request the examples' dependency closure explicitly (--filter='./examples/*^...') so the step no longer depends on which package happens to live at the top level. - ADR-0076: the round-2 status verification merged from main (#3061) was read against a pre-#3028 snapshot — svcAvailable no longer hardcodes 'available'; the D12 framework side shipped in #3028. Correct the status line and note OQ#9/OQ#10 resolution (#3037). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(ci): cap turbo test concurrency at the runner's core count (OOM mitigation) Test Core started failing deterministically on the hosted runner — plugin-audit#test dies with ZERO vitest output (kernel OOM-kill signature) while passing locally on Node 20 and 22 — first on main @4f8c2d1, then twice on this PR (initial run + job re-run). Onset correlates with the task graph growing past ~100 tasks (connector packages + qa gates): turbo's default concurrency (10) schedules that many vitest workers + tsup DTS builds onto a 4-vCPU runner and peak memory tips over. Cap test-step concurrency at 4 (PR affected-tests + push full-run). CPU-bound workload, so wall-clock cost is minimal; peak memory becomes bounded. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(qa): bump repo-relative path in the newly-merged declarative-mcp dogfood test + regen lockfile after merge showcase-declarative-mcp.dogfood.test.ts landed on main (#3062) against the old packages/dogfood location; after the packages/qa migration its '../../../examples/app-showcase' URL resolved inside packages/. One more level, same as the other conformance tests. Lockfile regenerated after taking main's side in the merge. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu --------- Co-authored-by: Claude <noreply@anthropic.com>
68 tasks
os-zhuang
added a commit
that referenced
this pull request
Jul 16, 2026
* feat(transport): prove the IHttpServer port is Hono-free — second adapter on raw node:http (ADR-0076 D11/OQ#10, #2462) Multi-adapter was designed but unproven: only the Hono adapter existed, and ADR-0076 D11 flagged 'the normalized context shows Hono-isms' as a caveat blocking the transport decomposition. This adds the validation: - New @objectstack/plugin-node-server: a thin IHttpServer implementation on raw node:http with ZERO dependencies beyond @objectstack/core — :param + trailing-* routing (registration-order first-match, same as the primary adapter), eager JSON/urlencoded body parse with lazy rawBody() for binary, SSE via native res.write/end, 404/405-with-Allow semantics, EADDRINUSE retry, graceful drain on close. Deliberately no getRawApp()/mount(). - Cross-adapter conformance suite: boots the dispatcher bridge AND the REST route generator (with ObjectQL + memory driver) on BOTH adapters over real sockets — /data CRUD roundtrip, /meta reads, /ready, /health, discovery, :param routing, 404/405 parity, plus a probe-for-probe response-shape parity matrix between the two adapters. 40 assertions, all green. Findings recorded in ADR-0076 (OQ#9/OQ#10/OQ#11 resolution notes): - The port has NO hard Hono-isms. The Host-header backfill in the Hono adapter is a Fetch-API artifact local to that adapter, not a port leak. - All remaining Hono coupling is confined to the feature-detected getRawApp() escape hatch (metadata HMR, cloud-connection routes, static/SPA + CORS + Server-Timing) which degrades gracefully. - Follow-up for the D11 window: codify res.write/res.end (SSE) and getPort() — extensions consumers already rely on — plus 404/405 semantics into the IHttpServer contract. - OQ#9 audit verdict (delineate, don't merge; only discovery + packages are genuinely double-mounted) recorded in the ADR open-questions list. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(plugin-node-server): harden query/params against prototype pollution (CodeQL) CodeQL flagged remote property injection: query-param keys come straight from the request URL, so building the map on a plain object literal let '?__proto__=…' write through the prototype chain. Use null-prototype objects for query and params and drop the dangerous keys (__proto__/constructor/prototype) outright; regression test pins that '?__proto__=polluted' is dropped and Object.prototype stays clean. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(plugin-node-server): eliminate the dynamic property-write sink CodeQL flags The blacklist guard added in the previous commit was not recognized as a sanitizer by CodeQL's js/remote-property-injection rule — the sink is the computed write itself (obj[userKey] = …). Build query and params via Object.fromEntries (own data properties only, no prototype-chain walk) so the sink no longer exists; the __proto__/constructor/prototype drop stays as defense in depth. Regression test unchanged and green (41/41). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * refactor(qa): reposition the port-conformance adapter as a private QA gate — packages/qa/http-conformance Maintainer review: the node:http adapter is a validation instrument, not a product server, so it must not ship as a publishable plugin- package. Reposition it accordingly: - Move packages/plugins/plugin-node-server → packages/qa/http-conformance; rename @objectstack/plugin-node-server → @objectstack/http-conformance, private: true, test-only (no build/publish), matching the existing verification-gate packages (dogfood, downstream-contract). - packages/qa/ becomes the unified home for non-published verification gates (new pnpm-workspace glob); migrating dogfood/downstream-contract there is proposed as a follow-up. - Drop the changeset + fixed-group entry (private packages are not versioned); update the ADR-0076 OQ#10 note to the new name. Behavior unchanged: same adapter, same 41-assertion cross-adapter conformance suite, all green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * refactor(qa): migrate dogfood + downstream-contract into packages/qa/ (unified verification-gate home) Completes the packages/qa/ consolidation started with http-conformance: all three non-published verification gates (dogfood regression gate, downstream-contract compatibility gate, http-conformance port gate) now live under one directory. Mechanical path migration — no behavior change: - git mv packages/{dogfood,downstream-contract} → packages/qa/ - rewrite every 'packages/dogfood' / 'packages/downstream-contract' breadcrumb: spec liveness proof registry + JSON ledgers, spec-liveness-check.yml path trigger, eslint ignore, docs/ADRs, source comments - bump the repo-root traversal one level in the four dogfood conformance tests that resolve REPO_ROOT from __dirname Verified from the new locations: dogfood 273 passed / 3 skipped, downstream-contract typecheck + 14 tests, spec check:liveness resolves all bound proofs, http-conformance 41/41. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * chore: retrigger CI — Actions did not fire on dc6507c Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * docs(changeset): fix dogfood path breadcrumb after packages/qa migration Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(ci): build the examples' dependency closure explicitly in the typecheck job + correct stale ADR-0076 D12 status Two follow-ups after merging latest main: - lint.yml: the example-app typecheck step built only ./packages/* (direct children). The connector packages the showcase imports were built purely by accident — through dogfood's dependency chain — which broke when dogfood moved to packages/qa/. Request the examples' dependency closure explicitly (--filter='./examples/*^...') so the step no longer depends on which package happens to live at the top level. - ADR-0076: the round-2 status verification merged from main (#3061) was read against a pre-#3028 snapshot — svcAvailable no longer hardcodes 'available'; the D12 framework side shipped in #3028. Correct the status line and note OQ#9/OQ#10 resolution (#3037). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(ci): cap turbo test concurrency at the runner's core count (OOM mitigation) Test Core started failing deterministically on the hosted runner — plugin-audit#test dies with ZERO vitest output (kernel OOM-kill signature) while passing locally on Node 20 and 22 — first on main @4f8c2d1, then twice on this PR (initial run + job re-run). Onset correlates with the task graph growing past ~100 tasks (connector packages + qa gates): turbo's default concurrency (10) schedules that many vitest workers + tsup DTS builds onto a 4-vCPU runner and peak memory tips over. Cap test-step concurrency at 4 (PR affected-tests + push full-run). CPU-bound workload, so wall-clock cost is minimal; peak memory becomes bounded. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * fix(qa): bump repo-relative path in the newly-merged declarative-mcp dogfood test + regen lockfile after merge showcase-declarative-mcp.dogfood.test.ts landed on main (#3062) against the old packages/dogfood location; after the packages/qa migration its '../../../examples/app-showcase' URL resolved inside packages/. One more level, same as the other conformance tests. Lockfile regenerated after taking main's side in the merge. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu * ci: TEMPORARY #3071 diagnostics — isolated verbose plugin-audit runs + dmesg OOM probe plugin-audit#test dies on the hosted runner with zero captured output (3x deterministic since main@4f8c2d1) while passing locally under Node 20/22 with the exact CI command. Bracket the turbo test step with an isolated, verbose, ungrouped vitest run (before: fresh machine; after: post-graph machine + kernel OOM traces) so the real failure output cannot be lost to the GH log pipeline. SIGQUIT-on-timeout dumps Node thread stacks if it hangs. To be REMOVED once #3071's root cause is fixed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A1BtxNeUaGmvUYr8FwtUNu --------- Co-authored-by: Claude <noreply@anthropic.com>
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
2 participants
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.
实现 #3056 的 showcase 演示部分:落地 ADR-0097 最初设想的
provider: 'mcp'验收演示(#3017 延后项),并用 dogfood 测试把它钉进 CI。前置安全闸 #3055(默认拒绝声明式 stdio)已随 #3059 合入,本 PR 是其第一个 dogfood 用户。变更
showcase:live
provider: 'mcp'演示(CI 确定性)scripts/mcp-fixture.mjs—— 仓内极小 stdio MCP 服务器,一个确定性工具echo_upper。选 stdio fixture 而非 http 自连:spawn 发生在 materialization 时刻,无网络、无端口、无启动时序耦合,Dogfood 门不会时序抖动。DevToolsMcpConnector(provider: 'mcp',transportnode ./scripts/mcp-fixture.mjs)+ShowcaseMcpConnectorEchoFlow(connector_action →echo_upper,输出变量携带structuredContent)。objectstack.config.ts接线new ConnectorMcpPlugin({ declarativeStdio: ['node'] })—— dogfood feat(connector-mcp): policy-gate declarative stdio transports — default-deny + host allowlist (ADR-0097 security follow-up) #3055 的 opt-in,注释如实说明粗粒度信任边界;coverage 注记同步(三个通用执行器的声明式实例现已全部演示)。dogfood:把演示钉进 CI
showcase-declarative-mcp.dogfood.test.ts启动真实 showcase(chdir 到应用根,与os dev的运行方式一致;vitest 每文件独立进程,chdir 隔离),经extraPlugins注入三个执行器(bootStack不装配 stack 的plugins:),断言:GET /automation/connectors列出showcase_mcp_tools:origin: 'declarative'、state: 'ready'、含echo_upper(即 fixture 成功 spawn +tools/list映射);rest/openapi 姊妹实例同样 ready;catalog descriptor 不在注册表;OBJECTSTACK(fixture 的tools/call结果完成全链路往返)。文档
content/docs/automation/flows.mdx节点表后新增一段:connector_action的连接器从哪来、三个通用执行器与声明式 provider 条目的关系(showcase 已全部接线)、stdio opt-in 的安全语义。验证
turbo test(dogfood + showcase + create-objectstack)63 任务全绿 ✓validate(23 flows / 8 plugins)+typecheck+ 55 测试 ✓;eslint ✓;lockfile 已更新(showcase/dogfood 新依赖)Refs #3056(showcase 演示部分;模板预设部分待 15.1.0 发布)、#3055 / #3059、#3017 / #3049、ADR-0097 §6。
🤖 Generated with Claude Code
https://claude.ai/code/session_01Pbmw3pMqNJfPbkvwh9Fkjs