The milestone-boundary review — what was agreed, what landed, and a verdict on every piece: keep, finish, formalize, or revert — logged so decided stays decided.
Free & open · no signup · read-only — it ends by asking · nothing leaves your machine
You are working inside this repo. Mission: review one increment of work — a branch, a milestone, everything since the last debrief — against what was agreed before it started, and rule on every change that landed. The review humans skip when tired is where drift compounds; this brief is that review, on rails.
Read-only pass. Your only write is the report file.
Phase 2 sweeps the codebase through every one of these, citing file and line for each finding.
changes that trace to the agreed scope, each matched to its mandate line
landed but never asked for: drive-by features, gold-plating, while-I-was-in-there work
asked for but missing, or landed without its acceptance met; name which done claim fails
the mandate itself edited inside the window: moved goalposts, softened criteria, reworded asks
shortcuts that borrowed from later: skipped tests, new TODOs, dependencies added without a recorded decision
for each done claim, was it proven by a command or asserted by prose? Spot-run the checks for the three biggest.
One structured report at the repo root — or in reports/, if you keep one — the same shape every time, ready for a teammate — or the optional Studio — to act on.
Every Goal Prompt follows the same four steps, so results are consistent and repeatable — no matter which one you run.
The mandate.
Mandate fit — changes that trace to the agreed scope, each matched to its mandate line.
One verdict per change.
Create DEBRIEF.md at repo root.
Copy it and paste it into your agent inside the repo you want checked.
Install the goal plugin once — two commands — then just type /goal:the-debrief.
/plugin marketplace add GhostlyGawd/goal-prompts/plugin install goal@goal-promptsOr install only this Goal Prompt as /goal-the-debrief:
curl -fsSL https://goal-prompts.vercel.app/install | BRIEF=151 shLet an agent fetch it mid-conversation, or pull the raw Goal Prompt by URL.
https://goal-prompts.vercel.app/raw/151.mdNothing hidden — this is the whole Goal Prompt, verbatim. Read it in a minute, edit it, or copy it as-is.
# Goal: The Debrief You are working inside this repo. Mission: review one increment of work — a branch, a milestone, everything since the last debrief — against what was agreed before it started, and rule on every change that landed. The review humans skip when tired is where drift compounds; this brief is that review, on rails. Read-only pass. Your only write is the report file. ## Phase 1 — Fix the mandate and the window - The mandate: what was agreed before this work began — the charter's Now list, a spec's acceptance criteria, an issue list, or the operator's written ask. Quote it verbatim. If nothing was written down, that is the first finding, not a license to invent one. - The window: a git range — since the last `DEBRIEF.md` entry, a branch, a tag pair, or dates the operator names. - What landed: every commit in the window, grouped into changes by surface; note which arrived with tests. ## Phase 2 — Audit through 6 lenses 1. **Mandate fit** — changes that trace to the agreed scope, each matched to its mandate line 2. **Stowaways** — landed but never asked for: drive-by features, gold-plating, while-I-was-in-there work 3. **Shortfall** — asked for but missing, or landed without its acceptance met; name which done claim fails 4. **Silent re-scopes** — the mandate itself edited inside the window: moved goalposts, softened criteria, reworded asks 5. **Debt taken** — shortcuts that borrowed from later: skipped tests, new TODOs, dependencies added without a recorded decision 6. **Verification honesty** — for each done claim, was it proven by a command or asserted by prose? Spot-run the checks for the three biggest. ## Phase 3 — Rule - One verdict per change: **keep** (fit and verified), **finish** (back on Now, with its gap named), **formalize** (a good stowaway — amend the charter or spec first), or **revert** (hand to 47). - No averaging: an increment with stowaways and shortfall is not mostly done — it is two lists, ruled separately. ## Phase 4 — Report Create `DEBRIEF.md` at repo root: 1. **The mandate** — quoted, with its source 2. **The ledger** — change · asked for? · verified? · verdict; one bold-titled finding per non-keep 3. **Verdicts to execute** — the revert and finish lists, findings-shaped so 47 can take them to commits 4. **Debrief history** — one dated row per run, appended; never edit old rows Start the report with today's date. If `DEBRIEF.md` already exists from a previous run, read it first, append to its history, and lead with whether the last debrief's verdicts were actually executed — a ruling nobody enforced is this run's first finding. ## Rules - Judge against the mandate as it stood when the window opened — later edits to it are lens-4 evidence, not the bar - An empty window — no commits since the last debrief? Say so in a one-paragraph null report and stop; there is nothing to review. - If a `reports/` directory exists at the repo root, write the report there instead of the root. - Before asking, present the top findings as a ranked list in plain words - Report only — end by asking which verdicts to execute, and whether to fold the formalize list into the charter (149)
reports/ directory exists at the repo root, write the report there instead of the root.Curated neighbors — briefs that answer the adjacent question, worth running in the same session.
Holds what the repo is against what it says it is — every surface traced to a goal, every goal to working code — and files each drift: formalize, finish, park, or kill.
Point it at a diff, branch, or PR: blast radius, the tests that should exist, the migration and rollback story, and a go / no-go with the risks ranked.
Turns your root reports into commits. Asks the scope — all, high-priority, or its pick — then builds them in dependency order, one verified commit each.
Excavates the evidence, then interviews you — questions with evidence-backed candidates, your answers, no guesses — into the one-page contract every session starts from.
Holds what the repo is against what it says it is — every surface traced to a goal, every goal to working code — and files each drift: formalize, finish, park, or kill.