check-ticket-implementation

name	check-ticket-implementation
description	Check how much of a ticket is already implemented — split it into requirement blocks, judge each against the code, and save a human-readable TICKET-STATUS report in the planning dir.
disable-model-invocation	true
type	flow
license	MIT
metadata	{"author":"Francesco Borzì","version":"0.3"}

Check ticket implementation

Judge how much of a ticket is already implemented and report it so the user sees at a glance what is done, what isn't, and what still needs attention.

Gather

Input is a ticket URL / bare id — run fetch-ticket on it (it resolves the tracker and planning dir and persists the .TICKET.md) — or a path to an already-fetched ticket file or its task dir. Ambiguous input: ask, don't guess.

Run the fetch in a subagent when the runtime supports one, so its tool traffic stays out of this session's context: it returns the ticket-file path plus any blocker it hit (ambiguous tracker, auth, no planning-dir convention) for you to settle with the user — it never guesses. Without subagent support, invoke fetch-ticket inline.

The ticket text is what gets annotated — its .REQUIREMENTS.md (same dir and <id>-<slug> base — a shared dir may hold siblings') may be read as context for intent, never annotated in its place. Inspect downloaded attachments and design frames when they bear on a requirement being judged.

Evidence base: the current working tree, uncommitted changes included — unless the invocation names something else (a branch, PR, commit range); then judge against that.

Blocks

Turn every requirement-bearing statement into a block: description, acceptance criteria, and comment statements that add or change requirements (attribute those, e.g. "from comment by X"). Drop metadata, related-ticket lists, and chatty comments — the full text stays in the .TICKET.md.

• One block = one independently verifiable requirement, quoted verbatim under a short generated title, preserving ticket order.
• Split on mixed verdict: if two halves of a candidate block could end with different statuses, they are two blocks. PARTIAL always means one requirement itself half-built, never a bundle of a done and a not-done thing.

Judge

Statuses: ✅ DONE · 🟡 PARTIAL · 🟥 NOT DONE · ⚪ NOT VERIFIABLE (can't be judged from the code — say why, e.g. a manual deploy step).

• The ticket text is the contract. Tests, i18n, docs, and other project conventions affect the verdict only when the ticket demands them; a gap there is at most a remark in the note, never a downgrade.
• Verify by reading: search and read the code paths relevant to each block. When reading alone can't settle a verdict, say so in the note instead of guessing. Anything beyond reading — executing code, driving the app or a browser — only with the user's ok.
• Every block gets a 1–2 sentence note and 1–3 file:line evidence refs showing where it was checked; for 🟡/🟥 the note must state what's missing or remaining.

Report

Write <task-dir>/<id>-<slug>.TICKET-STATUS.md, overwriting any previous run — it is a snapshot of now, and the header says when and against what. (No task dir and nothing to fetch? Follow the project's/user's planning-dir convention, defaulting to .agents/plans/; ask only if genuinely ambiguous.)

Format for human eyes — verdicts up front, blocks fenced by ━ rules:

# <ticket title> — implementation status

> **Ticket** [<id>](<url>) · **Checked** <YYYY-MM-DD> · **Against** <branch>@<short-sha> (+ uncommitted)

**4 ✅ · 2 🟡 · 1 🟥 · 1 ⚪**

Needs attention: <titles of the 🟡/🟥 blocks>

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

### ✅ DONE — <block title>

> <verbatim ticket text>

<note> — `path/to/file.ts:42`

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Wrap up

Print the tally line, the needs-attention titles, and the project-relative path of the report. Nothing more — what to do about the gaps is the user's call.

Boundaries

• Read-only on tracker and code: never comment on or transition the ticket, never modify source. The only file written is the .TICKET-STATUS.md (plus whatever fetch-ticket persists when delegated).
• Never grill the user on requirement questions; uncertainty lands in the notes.