Home/Catalog/Quality
Quality — does it work?
Brief 01

Bug Hunt

Trace latent bugs, unhandled edge cases, and race conditions; rank by user pain with repro paths and fix sketches.

BUGS.md4 phases8 lenses~2.7k chars

Free & open · no signup · read-only — it ends by asking · nothing leaves your machine

What it does

You are working inside this repo. Mission: find the bugs that are already here — latent, intermittent, or quietly shipping — and rank them by user pain. Not style issues, not refactors: defects.

Read anything, run read-only checks (tests, linters, the app itself). Don't fix anything yet. Your only write is the report file.

What it looks for

8 lenses it looks through

Phase 2 sweeps the codebase through every one of these, citing file and line for each finding.

1

Unhandled failures

missing try/catch on I/O, unchecked return values, promise rejections with no handler

2

Null paths

optional data dereferenced as if guaranteed; API responses trusted blindly

3

Race conditions

async ordering assumptions, stale closures, double-submits, missing debounce/locks

4

Boundaries

empty lists, zero, negative, huge inputs, unicode, max lengths, pagination edges

5

Time & math

timezone handling, off-by-one, float comparison, date arithmetic across DST

6

State drift

UI state vs server truth, cache invalidation gaps, optimistic updates that never reconcile

7

Validation gaps

client-only validation, type coercion surprises, trusting input shape

8

Leaks

listeners, subscriptions, timers, connections never cleaned up

The deliverable

What lands in your repo

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.

BUGS.md
1
Summary
count by severity; the single scariest finding
2
Findings
each: Name · Severity + confidence · Location (path:line) · Trigger (how a user hits it) · Expected vs actual · Root-cause hypothesis · Fix sketch (2–3 lines) · Effort S/M/L
3
Verification plan
fastest way to confirm each worth-verifying item
4
Top 3 to fix first
and why
One file. Evidence-backed. It ends by asking before touching anything. · see a real BUGS.md ↗
How it works

What it makes the agent do

Every Goal Prompt follows the same four steps, so results are consistent and repeatable — no matter which one you run.

1
Phase 1

Map the blast zones

Before hunting, learn where bugs hurt most.

2
Phase 2

Hunt through 8 lenses

Cite file and line for every suspect.

3
Phase 3

Curate

Keep only suspects you can argue from the code, with a plausible trigger scenario.

4
Phase 4

Report

Create BUGS.md at repo root.

Get started

Three ways to run this Goal Prompt

01 · COPY

Paste it in

Copy it and paste it into your agent inside the repo you want checked.

02 · INSTALL

As a slash command

Install the goal plugin once — two commands — then just type /goal:bug-hunt.

1/plugin marketplace add GhostlyGawd/goal-prompts
2/plugin install goal@goal-prompts

Or install only this Goal Prompt as /goal-bug-hunt:

curl -fsSL https://goal-prompts.vercel.app/install | BRIEF=01 sh
03 · AGENT

From an agent (MCP)

Let an agent fetch it mid-conversation, or pull the raw Goal Prompt by URL.

https://goal-prompts.vercel.app/raw/01.md
Transparency

The exact prompt

Nothing hidden — this is the whole Goal Prompt, verbatim. Read it in a minute, edit it, or copy it as-is.

Read the full brief (2,695 characters)
# Goal: Bug Hunt

You are working inside this repo. Mission: find the bugs that are already here — latent, intermittent, or quietly shipping — and rank them by user pain. Not style issues, not refactors: defects.

Read anything, run read-only checks (tests, linters, the app itself). Don't fix anything yet. Your only write is the report file.

## Phase 1 — Map the blast zones
Before hunting, learn where bugs hurt most:
- Which flows touch money, auth, or irreversible data changes?
- Where does state live, and who mutates it?
- Which code is newest, most complex, or most-edited? (check git log)

## Phase 2 — Hunt through 8 lenses
Cite file and line for every suspect.
1. **Unhandled failures** — missing try/catch on I/O, unchecked return values, promise rejections with no handler
2. **Null paths** — optional data dereferenced as if guaranteed; API responses trusted blindly
3. **Race conditions** — async ordering assumptions, stale closures, double-submits, missing debounce/locks
4. **Boundaries** — empty lists, zero, negative, huge inputs, unicode, max lengths, pagination edges
5. **Time & math** — timezone handling, off-by-one, float comparison, date arithmetic across DST
6. **State drift** — UI state vs server truth, cache invalidation gaps, optimistic updates that never reconcile
7. **Validation gaps** — client-only validation, type coercion surprises, trusting input shape
8. **Leaks** — listeners, subscriptions, timers, connections never cleaned up

## Phase 3 — Curate
- Keep only suspects you can argue from the code, with a plausible trigger scenario
- Severity: **S1** data loss / security / money · **S2** feature broken for some users · **S3** annoyance
- Confidence: certain / likely / worth-verifying

## Phase 4 — Report
Create `BUGS.md` at repo root:
1. **Summary** — count by severity; the single scariest finding
2. **Findings** — each: Name · Severity + confidence · Location (path:line) · Trigger (how a user hits it) · Expected vs actual · Root-cause hypothesis · Fix sketch (2–3 lines) · Effort S/M/L
3. **Verification plan** — fastest way to confirm each worth-verifying item
4. **Top 3 to fix first** — and why

Start the report with today's date. If `BUGS.md` already exists from a previous run, read it first and lead with what changed since.

## Rules
- Evidence or it doesn't exist: every finding cites code
- A trigger scenario is required — "this looks wrong" isn't a bug report
- Depth on real defects beats volume of nitpicks
- 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 bugs to fix

House rules for this Goal Prompt

  • Evidence or it doesn't exist: every finding cites code
  • A trigger scenario is required — "this looks wrong" isn't a bug report
  • Depth on real defects beats volume of nitpicks
  • 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 bugs to fix
Keep exploring

More Quality briefs

Runs inside these playbooks — curated sequences you can launch with one paste: