/app-status

description: Show the current state of the project — vision, sprint goal, board, blockers, daily report
allowed-tools: Read, Glob, Grep, Bash

/app-status — Where are we?

Steps

1. Print the project's one-line vision from `docs/00-vision.md` (first heading + first paragraph). 2. Print the current sprint goal from `docs/30-sprint-plan.md`. 3. Print the board summary from `docs/31-board.md`:

• Count of tickets per status (todo / in_progress / review / qa / done / blocked)
• List any `blocked` rows or rows with notes
• **Run the board doctor** (`board-doctor` skill) and print its verdict:

     node "${CLAUDE_PLUGIN_ROOT}/scripts/board-doctor.mjs" docs/31-board.md

Anomalies go at the **top** of the status output, not the bottom. A `stranded` ticket is the single most important thing on this screen: it is work the sprint loop will never surface on its own. 3aa. **Say which `qa` rows have actually landed.** `qa` now covers two states and they need different answers:

   node "${CLAUDE_PLUGIN_ROOT}/scripts/merge-reconcile.mjs" --root .

• `AWAITING INTEGRATION: …` → those tickets passed the merge gate and the wave has not run. Print

them as **"gated, not yet landed"** with the command that lands them (`wave-integrate.mjs --wave <N>`). This is the loop working, not a fault.

• `BLOCKED — claims integrated code that is NOT in <base>` → a ticket says it shipped and git

disagrees. That goes at the **top** of this screen next to the doctor's anomalies.

Without this, a founder reading `4 in qa` cannot tell "four features are on the integration branch" from "four features are approved and still sitting on their own branches".

3a. **Render the board** so the shape is visible, not tabular:

   node "${CLAUDE_PLUGIN_ROOT}/scripts/board-render.mjs" docs/31-board.md --out docs/32-board-view.md

Print the kanban, the per-owner load, and the NEEDS ATTENTION block. **`--out` is not optional:** without it, checking status never refreshes the committed `docs/32-board-view.md`, which is the view humans actually read on GitHub — it silently drifts away from the board it claims to show.

3b. **Self-metrics — the only quantitative thing this system can say about itself.** Derived from `docs/31-board-events.jsonl` by `lib/events.mjs`; read them out of the CLI:

   node "${CLAUDE_PLUGIN_ROOT}/scripts/board.mjs" show --json

Print the `metrics` object as exactly this block, and nothing more:

   SELF-METRICS  (E events, I inferred)
     cycle time (median)   4h 12m   claimed -> closed, over T tickets that have both
     review pass rate      67%      4 of 6 tickets approved with no rework
     rework rate           33%      2 of 6 went back for changes
     gates fired           rejected 1 · changes 2 · qa_failed 0 · blocked 1
     tickets per round     2026-07-28 3 · 2026-07-29 2

Rules, because a metric that lies is worse than no metric:

• `reviewPassRate` and `reworkRate` are `null` until some ticket has reached review. Print

**`n/a`**, never `0%` — an empty denominator reads as "every review failed".

• `medianCycleTimeMs` is `null` when no ticket has both a `claimed` and a `closed`/`merged`

timestamp. Print `n/a — no ticket has completed`, and say how many are in flight.

• **Migrated events have `ts: null`**, so a project that was migrated has tickets that cannot

contribute a cycle time. Print the inferred count in the header and, if it is more than half the log, add one line: `most of this log is inferred from a hand-written board — timings start from the migration, not from the sprint`. A thin number presented as a measurement is the failure mode this block exists to avoid.

• `gateFires` counts times a gate **caught** something. All zeros on a busy board is not a

healthy team; it is a loop that is not running its gates. Say so rather than printing it approvingly.

• Exit `2` → the log is missing or unreadable. Print `SELF-METRICS: CANNOT EVALUATE` with the

reason. Exit `1` with output → the log folds with sequence violations (hand-appended lines); print the metrics *and* the violation list, since the numbers are derived from a log that disagrees with itself.

• No `docs/31-board-events.jsonl` at all → print `SELF-METRICS: n/a — this project's board is

hand-written; run /app-plan or /app-build once to migrate it`. Do not migrate from here: `/app-status` is a read-only command and migration rewrites `docs/31-board.md`.

3bb. **Loop trend and budget** — what happened to the *loop*, which the event log cannot answer:

   node "${CLAUDE_PLUGIN_ROOT}/scripts/round-journal.mjs" show
   node "${CLAUDE_PLUGIN_ROOT}/scripts/round-journal.mjs" check

Both read `docs/33-rounds.jsonl`, appended once per round by `/app-build` step 6a. `show` is the per-round burn-down (tickets waved, verdicts, retries, refusals); `check` is the one-line budget position, and it prints here whether or not it is near a ceiling — a spend first seen when it stops the run was seen too late. `ROUND JOURNAL: no rounds yet` is the correct output before the first `/app-build` round; it is not an error.

**Spend reads `not measurable in this harness` unless a round actually reported a number.** That sentence is the honest one: rounds, spawns, retries and refusals are counted because they are countable, and no token figure is invented to fill the gap.

3c. **Open team threads.** From `docs/team/messages.jsonl`, list any `question` with no matching `answer`, and any `escalation` not yet closed by a `decision`. These are the conversations the team is stuck inside.

4. Print today's daily report if it exists at `docs/daily/<today>.md`. 5. Print recent bug list from `docs/51-bugs.md` if it exists — show open S1/S2 only. Then print the register view at `d