How user signal re-enters the product — corrections captured or evaporating, thumbs nobody reads, and the complaint-to-fix path.
Free & open · no signup · read-only — it ends by asking · nothing leaves your machine
You are working inside this repo. Mission: trace how user signal enters this product and whether it ever changes anything — or whether feedback is collected into a void.
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.
the richest signals uncollected: a user editing the agent's output is a labeled correction, vanishing on save; a retry is a thumbs-down nobody logged
thumbs stored in a table with zero readers; the census of write-only feedback
complaint → fix: is there a path at all, and how long is it in practice
bad runs becoming eval cases (34), corrections becoming few-shots or prompt fixes (30), feature pain becoming roadmap items
"you reported this, it's fixed" — the cheapest retention message that almost nobody sends
does any human regularly read real transcripts, or only dashboards of averages
rating prompts so frequent that response rates collapse; measure what's actually answered
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.
Every way users tell you something.
Capture gaps — the richest signals uncollected.
Rank channels by signal quality × current waste.
Create FEEDBACK.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:feedback-loop-audit.
/plugin marketplace add GhostlyGawd/goal-prompts/plugin install goal@goal-promptsOr install only this Goal Prompt as /goal-feedback-loop-audit:
curl -fsSL https://goal-prompts.vercel.app/install | BRIEF=45 shLet an agent fetch it mid-conversation, or pull the raw Goal Prompt by URL.
https://goal-prompts.vercel.app/raw/45.mdNothing hidden — this is the whole Goal Prompt, verbatim. Read it in a minute, edit it, or copy it as-is.
# Goal: Feedback Loop Audit You are working inside this repo. Mission: trace how user signal enters this product and whether it ever changes anything — or whether feedback is collected into a void. Read-only pass. Your only write is the report file. ## Phase 1 — Map the signal channels - Every way users tell you something: thumbs and ratings, edits and corrections of outputs, retries, abandonments, support tickets, cancellations. - For each: where does the signal land — a table, a tool, a Slack channel, nowhere? - Find the readers: search the code and docs for anything that *consumes* each signal. ## Phase 2 — Audit through 7 lenses 1. **Capture gaps** — the richest signals uncollected: a user editing the agent's output is a labeled correction, vanishing on save; a retry is a thumbs-down nobody logged 2. **Dead-end signals** — thumbs stored in a table with zero readers; the census of write-only feedback 3. **Loop latency** — complaint → fix: is there a path at all, and how long is it in practice 4. **Signal-to-artifact wiring** — bad runs becoming eval cases (34), corrections becoming few-shots or prompt fixes (30), feature pain becoming roadmap items 5. **Closing with users** — "you reported this, it's fixed" — the cheapest retention message that almost nobody sends 6. **Raw-session review** — does any human regularly read real transcripts, or only dashboards of averages 7. **Ask fatigue** — rating prompts so frequent that response rates collapse; measure what's actually answered ## Phase 3 — Curate - Rank channels by signal quality × current waste: high-quality ignored signals first - Every wiring proposal names source, destination artifact, and owner ## Phase 4 — Report Create `FEEDBACK.md` at repo root: 1. **Signal inventory** — channel · volume · destination · readers found 2. **Dead-ends ranked** — the write-only list 3. **The wiring plan** — signal → artifact (eval, prompt, roadmap) → mechanism 4. **The loop to close this week** — one signal, end to end, spelled out 5. **Ask hygiene** — what to stop asking Start the report with today's date. If `FEEDBACK.md` already exists from a previous run, read it first and lead with what changed since. ## Rules - Feedback that changes nothing is a survey, not a loop - Corrections are gold: captured labels users paid you with their patience to create - No user feedback channels in this repo? Say so in a one-paragraph null report and stop — a null result is a valid finding. - 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 loop to close first
reports/ directory exists at the repo root, write the report there instead of the root.The flagship. Understand the whole product, then surface prioritized opportunities across UI, UX, retention, engagement, community, and beauty.
Every message the product sends out of the app — transactional email, push, in-app — for whether it is timely, wanted, clear, and controllable.
The product's own search — whether users can find what they are looking for: relevance, recall, forgiveness, and what happens on zero results.
The whole arc of an account beyond signup — settings, export, plan changes, and deletion — and whether the product treats users with dignity, even leaving.