build(chat): consume partial-markdown 0.5.3 — streaming table body rows

[blove]

Bumps @cacheplane/partial-markdown to 0.5.3 in libs/chat: while a table is streaming, an in-progress body row now renders as a streaming row inside the table (growing cell-by-cell, padded to header width) instead of momentarily closing the table and flashing as a raw-pipe paragraph (bare | / one pipe) or a spurious second header-only table (two+ pipes). Completes the streaming-table work begun in 0.5.2 (#739) and the finalize fix (#743). Library side: cacheplane/cacheplane#22, tag partial-markdown-v0.5.3.

• New consumer anti-flicker test in streaming-markdown.table-stream.spec.ts — red on 0.5.2 (raw-pipe <p> at '|', 2 tables at '| x1 | y'), green on 0.5.3.
• Surgical lockfile edit (3 spots, no regeneration). No chat component changes.

Chrome MCP live verification (required gate) — PASSED

Frame-sampled the DOM (MutationObserver, per-mutation) during live gpt-5-mini table streams on examples-chat (vite dep cache cleared and confirmed serving the 0.5.3 module). Across 6 streamed-table runs:

• 0 frames with a second <table> (the 0.5.2 spurious-table artifact is gone)
• 0 frames with a raw-pipe <p> alongside the table (the body-row paragraph flash is gone; pre-fix baseline showed 2 such frames)
• Intact-content runs render table-every-frame from the first closed header cell; final tables correct (e.g. 4 cols × 12 rows).

Orthogonal finding (not this PR; filed separately)

The live gate exposed an intermittent, pre-existing streaming-accumulation bug: streamed deltas occasionally drop characters (captured: delimiter accumulated as ---:---:| while the final message had |---:|---|---:|), so the mid-stream content is invalid GFM and correctly renders as text until the final snapshot replaces it. Replaying the corrupted progression through the parser reproduces the paragraphs; clean content renders a table — proving it's upstream of partial-markdown. Tracked as a follow-up task.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
threadplane	Ready	Preview, Comment	Jul 6, 2026 3:17am

[github-actions]

Automated approval: this PR received an intelligent (AI) code review. See the review comments on this PR.

[github-actions]

Claude finished @blove's task in 2m 33s —— View job

---

Code Review

• Gather context
• Read CLAUDE.md and repo guidelines
• Read PR diff and changed files
• Analyze test coverage
• Review lockfile changes
• Post findings

---

Summary

This is a clean, minimal consume-PR: version bump in libs/chat/package.json, a surgical 3-spot lockfile edit, and one new anti-flicker test. No Angular component changes. The approach matches the stated design (projection-only fix in the library, consumer test as a red→green regression guard). Overall: LGTM with two minor notes.

---

Findings

1. Dead grow() call before the loop (minor — see inline comment)

streaming-markdown.table-stream.spec.ts line 103 calls grow('| A | B |\n| - | - |\n') before the loop, but the first iteration immediately overwrites the content — no assertion fires between the two. Either drop it or add an assertion to make the intent explicit (e.g. verify the committed header renders correctly before the partial-row steps begin). Fix this →

2. Test asserts absence of artifacts but not presence of body-row content (minor)

The loop checks table count === 1 and no raw-pipe <p>, which is exactly the right anti-flicker guard. However, for steps like '| x1 | y', the spec calls for the partial row to appear inside the table's <tbody>. A complementary positive assertion like el.querySelectorAll('tbody tr').length >= 1 at those steps would confirm the row content is actually projected (not just that the second table disappeared). Low priority — the library-side unit tests own the structural output; this consumer test's primary role is the anti-flicker gate. But it would make the test self-documenting about what the "fix" actually produces.

3. Lockfile integrity hash — verification note (informational)

The surgical lockfile edit correctly updates all three spots (root deps, libs/chat block, node_modules resolved block). The new SHA-512 for 0.5.3 (sha512-XlvjGVsEY7tA3Q4eSY0EiP3FmKkzv5VwFWCmmqvnvf1/dYy+YjQXmDg1mqtFWk4iNyjMMd7aBtGiY+lEuZPmLQ==) has the correct length and format. I wasn't able to run npm view @cacheplane/partial-markdown@0.5.3 dist.integrity in this environment to cross-check it against the registry, but the PR author's documented process (pack-and-replace from the live npm tarball) is the right approach. A reviewer with npm access can do a quick npm view @cacheplane/partial-markdown@0.5.3 dist.integrity to confirm.

---

No issues found in

• libs/chat/package.json: range bump ^0.5.2 → ^0.5.3 correct, no other deps affected.
• Lockfile: exactly 3 spots changed as claimed, no stray regeneration artifacts, node count intact.
• Doc files: plan/spec docs are agentic work artifacts, not public docs — their local repo paths (~/repos/cacheplane) are expected in that context.
• Public API: no changes to @threadplane/chat's exported surface.
• Angular/RxJS: no component changes, no subscription or change-detection concerns introduced.

[github-actions]

This initial grow is dead code — the first loop iteration immediately replaces the content with '| A | B |\n| - | - |\n' + '|', so no assertion fires against this intermediate state. You could either drop it or add an assertion here (e.g. expect(el.querySelectorAll('table').length).toBe(1)) if the intent was to verify the committed header+delimiter state before the partial rows begin.

Suggested change

	grow('| A | B |\n| - | - |\n');
	grow('| A | B |\n| - | - |\n');
	expect(el.querySelectorAll('table').length, 'committed header renders as table').toBe(1);