chore(examples/rsc-agent-runtime): drop skills boilerplate, add validate script, document why the empty target sections stay - #65
Conversation
…ate script, document why the empty target sections stay
|
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
commit: |
…n every test budget (#70) The dev-invocation and dev-provider polling helpers computed their deadlines from raw milliseconds, so the documented time-scale knob widened neither the 4s default nor the explicit per-call budgets, and the example's rstest config never raised the 5s default test timeout. On a contended host that combination expired real work instead of bounding a hang, which is the flake shape the PR #65 report described. Budgets now stay unscaled at the call sites and are scaled inside the helpers, matching state-and-definition.test.ts. Per-test timeouts, the shared default, and the capacity-fixture worker lifetime scale with the same knob; every value is unchanged when the knob is unset.
1 participant
Summary
Wave A3 of the examples refresh plan (audit §5, executable-now items) — scoped to
examples/rsc-agent-runtimeonly.skills: []fromagent-bundle.config.ts. The example has noskills/directory, so the empty list was boilerplate that post-RFC-RFC: framework mode — structure in config and conventions, JSX only for rendering #63 would read as a deliberate opt-out it isn't. Inspect parity verified: with only this drop,agent-bundle inspect --jsonbefore/after is identical inmodel,plans,diagnostics,state, andmodelDigest(onlyconfigDigest/revisionmove, tracking the config file bytes).claude: {}/codex: {}/portable: {}sections, per the audit's verify-first instruction. They are not redundant withtargets:: dropping them emptiesmodel.extensions— normalization materializes each section as anextension:<target>entry with config provenance, andmodelDigestchanges. Plans/diagnostics stay identical, but the inspect model does not, so the audit's parity bar fails and the sections stay, now with a config comment explaining why.validatescript (agent-bundle validate) topackage.jsonfor cross-example parity (verified: "Validation succeeded", exit 0). Thecheck-chain shape is left untouched for the Wave A4 consistency sweep.tests/docs-contract.test.tsneeds no lockstep change; no files moved, sodocs/architecture/rsc-runtime-workbench.mdneeds no regeneration (pnpm check:runtime-topologypasses).Owner decision requested (flagged, not fixed)
This example's
README.mdself-declares (line 3) that it is a "private, opt-in… architecture experiment, not anagent-bundlepublic API", and the body is dense with internal jargon ("skip-gated", "schema-v2 evidence envelope", "value-free hook launch probe"). That is in tension with the repository rule thatexamples/*are user-facing products; the root README's examples table already omits this example. Two ways to resolve it: (1) promote the example to product-quality documentation so it meets theexamples/*bar, or (2) formally mark it internal — carve out an explicit exemption in AGENTS.md or relocate it outsideexamples/— so the rule and the tree agree. This PR deliberately does not change the example's status either way; the choice is left to the owner.Test plan
pnpm --filter @agent-bundle/rsc-agent-runtime-demo check— 0 failures, 6 skips (the six Windows-only Job Object tests; known platform skips)pnpm check:runtime-topologypnpm eval:spotpnpm typecheckandpnpm lintagent-bundle inspect --jsonbefore/after parity check (see above)