Curate a batch of open issues with the good first issue label ahead of 2026-09-02 student session

[s-annam]

Problem

The good first issue label exists on offlinecv/OfflineCV and has been used before, but every issue currently carrying it is closed — there are zero open issues labeled good first issue right now. As new contributors come onboard, we should label eligible issues so they're visible and offer a low-risk on-ramp when people land on the issue tracker.

Ask

• Go through currently-open issues and apply good first issue to a handful that are genuinely scoped for a first-time contributor (small, well-isolated, clear acceptance criteria, no deep parser-cascade context required).
• Keep the bar real — mislabeling something that turns out to need cross-file context or deep parser knowledge burns a new contributor's first attempt.

Acceptance criteria

• At least a handful of currently-open issues carry good first issue.
• gh issue list --repo offlinecv/OfflineCV --label "good first issue" --state open returns a non-empty list before the 2026-09-02 session.

[s-annam]

Curated. Six open issues now carry good first issue, ordered here easiest first — a contributor can pick by appetite rather than by scroll position.

#	What it is	Why it clears the bar
#821	cleanRewriteLine eats the minus off -5% churn	One regex in one file. The issue carries the exact replacement, four repro lines, and names the it.each table the new case joins. Nothing to design.
#749	Same-millisecond savedAt sorts the résumé library oldest-first	One comparator at resume-library.ts:156. Root cause proved (stable sort + 0), two fix options with a recommendation. Teaches deterministic-test discipline without needing it to be discovered.
#902	CONTRIBUTING.md: link the GFI query, document comment-to-claim	Docs only, two acceptance criteria. Was blocked on this issue — the list it links is non-empty as of now, so it is unblocked.
#817	Two coverage gaps left by #682	Test-only, no production risk. Both fixes are spelled out and both are mutation-checked, so "did I actually fix it" has an objective answer.
#916	Shared lexicon (b): export the date tokens, delete 7 copies	The issue nominates itself — behaviour-preserving and mechanical. Byte-identical corpus snapshots are a hard, checkable stop condition.
#861	Split the four section renderers out of ReconstructedResume.tsx	The stretch pick. Pure move, zero behaviour change, and the target shape already exists in two sibling files. Largest diff of the six — worth saying so to whoever claims it.

What was rejected, and why

Keeping the bar real cuts both ways, so the near misses are recorded rather than left to be re-litigated:

• [edit] Confirm before deleting an entry — a one-click destructive delete that skips the confirmation its own bullets have #860 (confirm before deleting an entry) — carries a written Reuse analysis, but the work is lifting [edit] no way to delete a bullet outside the WebGPU-gated rewrite panel; cleared bullets leave a ghost row #626's pending-confirm shape across four call sites inside the repo's largest known-debt component. Cross-file UI state.
• preserve-numbers: dropped year masked by an unrelated same-value digit #876 (preserve-numbers year gate) — the issue itself asks for a false-positive audit against a design docblock. Judgement, not a patch.
• Eval rubric scores literal markdown as clean — add a residual-markdown criterion so a future run surfaces the class #805 (eval rubric residual-markdown criterion) — the body says outright it is not a small change: seven files and four open design questions.
• CI never runs Node 22, so the engines range #802 declares is unverified #808 (Node 22 in CI) — it is a CI-spend tradeoff plus a branch-protection change. A newcomer cannot land the second half.
• Ship an og:image social card so shared offlinecv.org links render a card (split from #753 Step 6) #774 (og:image card) — needs a designed 1200×630 asset and verification against the live origin after a release promote.
• [parser] skills — the final wrap of a comma-separated list never rejoins, shredding the last multi-word skill #834, [export] Non-bullet description content is dropped from the downloaded PDF #844, edit: a restored draft forgets the parked end date, so #814's loss returns across that boundary #819, [export] the one-anchor collapse also flattens the embedded resume.json, which could have preserved a lone end date #816 — all need the parser-cascade or export-seam context the label is supposed to exclude.
• LlmEscapeHatchPanel's mount-across-done invariant is asserted in three docblocks and tested nowhere #687 (mount-across-done invariant) — test-only and well-written up, but settling it needs React mount/unmount semantics a first-timer is unlikely to have.

Acceptance criteria

• At least a handful of currently-open issues carry good first issue — six.
• gh issue list --repo offlinecv/OfflineCV --label "good first issue" --state open returns a non-empty list before the session.

Downstream: #902 is unblocked, and the session slide that sends students to the filtered query now points at something.

[s-annam]

Correction — #749 pulled from the batch.

It was already fixed on main: src/lib/resume-library.ts:212 carries the deterministic tie-break (|| a.id.localeCompare(b.id)), landed under #907, which was filed and closed independently on 2026-08-27 while #749 stayed open. A newcomer would have opened it and found nothing to do. Label removed, #749 closed as a duplicate of #907.

The other five were re-verified against origin/main rather than trusted from their bodies:

#	Check	Still live
#821	LIST_MARKER_PATTERN at post-process.ts:113 still has the [•\-]\s* branch	yes
#817	no direct test references formatExperienceDateRange anywhere under src/lib/edit/	yes
#916	MONTH, SEASON, STRICT_MONTH still module-private in regex.ts (129/133/144)	yes
#861	file is now 1786 LOC, not the 1489 the body cites — it has grown a further ~300 since filing	yes, and larger
#902	CONTRIBUTING.md still has no mention of good first issue	yes

Five labelled issues, all verified against the tree today. Reading an issue body is not evidence the defect is still open — that is the check this batch nearly shipped without.