Repository navigation
docs: mirror the text-only review context note in the zh README and re-record the i18n pair - #163
Conversation
…e-record the i18n pair
✅ Verdict:
|
| Findings | Count |
|---|---|
| 🔴 must-fix | 0 |
| 🟠 should-fix | 0 |
| 🟡 may-fix | 0 |
| 🔵 nit | 0 |
📋 Review
What this PR does: Docs-only follow-up to #162 that closes the bilingual-pair gap it left behind. README.zh.md gains a 审阅上下文仅含文本 bullet in ## 限制与路线图, restoring the accepted-gap list to nine entries in the same order as the English side, and README.i18n.yaml re-pins both README.md and README.zh.md blob hashes so the consistency record describes the commit that actually made the pair consistent. No code, test, or typecheck surface is touched.
Findings
none
Revalidated findings
none
Linked-issue AC
none — the PR links no issue (empty issues list); it is a docs follow-up to PR #162 / issue #161 and states no acceptance criteria of its own to score.
✅ Verified
- Review document: ship it, 100% (mergeable), must-fix=0 should-fix=0 may-fix=0 nit=0, empty findings — nothing survives on either in-scope file.
- README.i18n.yaml in the tree pins README.md: 5bfd4e87… and README.zh.md: 8449c116…, matching the two hashes the PR description claims; the header comment and re-record procedure are unmodified.
- Both ## Limitations & roadmap / ## 限制与路线图 lists read as nine bullets in the same order, with the new text-only entry last, immediately after the bounded-context entry.
- The zh bullet carries the three claimed semantics: placeholders for image and file blocks, attachment bytes never reaching the reviewer, and primary-agent-read content still reviewed.
- ❓ git hash-object was not re-run in this review; the pins are consistent by inspection of the manifest, not by recomputation.
- ❓ No pnpm typecheck, test run, or docs build is evidenced; the docs-only scope rests on the PR description and the two files in scope.
🗑️ Considered & rejected
- Stale hash pins / manifest left describing the pre-#162 state — rejected: the manifest in the tree carries the post-#162 values for both sides, and its header comment and procedure are untouched as claimed.
- New zh bullet out of step with the en list — rejected: it is the ninth entry on both sides, in the same position (last, after the bounded-context gap), so the lists match entry for entry.
- Zh wording drift from the English bullet — rejected: the zh text names media type alongside the attachment digest and dimensions, which tracks the zh settings-card textOnlyContextHint; all three substantive points (placeholders only, no attachment bytes, primary-agent-read content still reviewed) resolve identically.
- as unknown as ContentBlock cast in tests/transcript.test.ts — rejected for this PR: it is pre-existing test code outside the in-scope docs files, and the PR description explicitly defers it to the test's owner rather than widening a docs-only change.
- Missing changelog/version note for the docs edit — rejected: this repairs a bilingual-pair inconsistency introduced by #162 with no user-facing behaviour or API change, so no release note is warranted.
🛠️ Plan to fix
Expand fix plan
none — no findings remain; the two-file docs change is consistent as committed and mergeable at 100%.Review 1 · commit b79c11b
Follow-up to #162 (issue #161).
Problem
#162 added the English
Text-only review contextbullet to## Limitations & roadmapbut did not:README.zh.md, so the zh accepted-gap list had 8 entries against the en list's 9;README.i18n.yaml, whose pinnedREADME.mdblob was still the pre-fix: render image/file attachments as identity placeholders in the review context #162 value (4b9a16be…) — the bilingual-pair record described a state that no longer existed.Change
README.zh.md: one bullet in## 限制与路线图, immediately after**advisor 上下文有界**, preserving list order and——style, worded to match the zhtextOnlyContextHinton the settings card: placeholders only, no attachment bytes, primary-agent-read content still reviewed, no image passthrough.README.i18n.yaml: both pinned blob hashes re-recorded; header comment and procedure untouched.Docs only — no code, test, or typecheck change.
Verification
git hash-object README.md→5bfd4e8750a1ba62e732d87ae25a16daf102aeb4;git hash-object README.zh.md→8449c116f6da8bf6221a3687e0028a9b57f4729a. Both match the recorded values byte-for-byte (and the manifest is inside this commit).Follow-up note (no action taken here)
Noted for whoever owns
tests/transcript.test.tsnext: the new image-only case builds its attachment withas unknown as ContentBlock. Tightening it to a real V4ContentBlockimage arm would assert the attachment shape instead of casting it away, and onepnpm typecheckwould confirm the four attachment field reads against the@deepseek-ai/dsh-llmtypes. Deliberately out of scope here — this PR is docs-only so the recorded hashes describe exactly the commit that made the pair consistent.