Tianyi Cui
75b97e7463
The Decision and Risks paragraphs described the change relative to the patch that landed it (what was updated together, what another change took first). An implemented RFC reads as standing truth: the pastes match the pruned maps, the manifest rows remain because the symbols survive, and the image block's cache field belongs to the drop-image RFC's record.
19 lines
2.3 KiB
Plaintext
19 lines
2.3 KiB
Plaintext
(A) merge-blocker
|
|
- `docs/rfc/implemented/simplification/2026-07-04-prune-producerless-vocabulary-variants.md:15` and `docs/rfc/implemented/simplification/2026-07-04-prune-producerless-vocabulary-variants.md:31` still narrate this PR/change history instead of just the shipped/current state: "Updated in the same change", "`scripts/type-equiv.manifest.json` is untouched", "shipped first", and "this change removed". The review spec asks for current-state prose only, and the house rules prohibit process narration in docs/comments. Reword these as present facts about the live vocabulary/catalogs rather than mechanics of this patch.
|
|
|
|
(B) non-blocking
|
|
- None.
|
|
|
|
(C) nitpick
|
|
- None.
|
|
|
|
Evidence checked
|
|
- Producer-less claims are accurate: exact `cache:` block constructors are gone except this RFC record (`rg "\bcache\s*:"`), exact `{ kind: 'agent' }` constructors are gone (`rg "\{\s*kind:\s*['\"]agent['\"]\s*[,}]"`), and exact `{ kind: 'continuation' }` constructors are gone (`rg "\{\s*kind:\s*['\"]continuation['\"]\s*[,}]"`).
|
|
- The loop constructs `message` at `packages/core/agent-loop/src/loop.ts:284` and `injection` at `packages/core/agent-loop/src/agent.ts:145`; production `trigger.kind` readers are the ACP message filters at `packages/ui/acp/src/index.ts:352` and `packages/ui/acp/src/index.ts:385`.
|
|
- The subagent prompt path sends without a source at `packages/subagent/subagent-inprocess/src/index.ts:155`, and missing sources default to `{ kind: 'user' }` at `packages/core/agent-loop/src/agent.ts:103`.
|
|
- The core-data-structures pastes no longer include the removed variants: `docs/core-data-structures/core.md:107` and `docs/core-data-structures/session.md:165`.
|
|
- The llm-replay fixture still proves `deriveReplayScript` ignores non-`assistant/chunk` events: the test includes `turn/start`/`turn/end` around chunks at `packages/support/llm-replay/tests/llm-replay.spec.ts:118`, while the implementation skips all non-chunk events at `packages/support/llm-replay/src/index.ts:195`.
|
|
- Checks run: `pnpm exec vitest run packages/support/llm-replay/tests/llm-replay.spec.ts` (51 passed), `pnpm run doc-sync` (passed), `git diff --check origin/simpl-a1-drop-image...HEAD` (passed), precise EOF check on touched files (passed), and `pnpm run typecheck` (passed).
|
|
|
|
Verdict: NO-SHIP as it stands, due to the implemented RFC process-narration blocker above.
|