Skip to content
Agentmart
Submit a tool

investigate

Investigates a bug, pasted error, "stopped answering", failed booking or customer conversation.

Skill by stage2dev

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/stage2dev-investigate

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Investigates a bug, pasted error, "stopped answering", failed booking or customer conversation. Read-only first, cause with proof, smallest fix. Use for "find out why", "investigar", "descobre o porquê".

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/investigate

Tools it pre-approves 4

From allowed-tools in SKILL.md: the agent may use these without asking you each time.

  • Bash(sh
  • "${CLAUDE_PLUGIN_ROOT}/scripts/py"
  • scripts/guarda.py
  • info)

What's inside

Skill name
investigate
Description
Investigates a bug, pasted error, "stopped answering", failed booking or customer conversation. Read-only first, cause with proof, smallest fix. Use for "find out why", "investigar", "descobre o porquê".
License
MIT
Allowed tools
Bash(sh, "${CLAUDE_PLUGIN_ROOT}/scripts/py", scripts/guarda.py, info)
Instructions
37 lines in SKILL.md · view on GitHub

Before installing

  • Pre-approves some shell commands

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Investigate - find the cause before touching anything

Context: !`sh "${CLAUDE_PLUGIN_ROOT}/scripts/py" scripts/guarda.py info`

Symptom: $ARGUMENTS

The JSON above is the context. Write to the user in its `language`, honoring `reply_style`; final answer short. Use document names from `names`, never literal. Models/effort per role from `models`; if missing use the defaults below.

## Rules

- First pass is READ ONLY: no edits, no re-enabling, no production writes, no customer messages.
- Read the project profile (`profile_path`): "diagnostic sources", production, pitfalls.
- If a project module has its own investigation skill (for example a conversation one), use it in place of step 3.
- A real customer is never a test subject.
- Secret or personal data seen in a log: do not repeat it in the answer, a file or memory.

## Method

1. **Identifier and window.** Get what identifies the case (phone suffix, id, e-mail, request) and the time window in the project's time zone. Missing one: ask ONCE, together with the partial answer you already have.
2. **Triage.** P0 = customer affected now or today; P1 = affects several, no urgency; P2 = pre-existing or cosmetic. Do P0 first.
3. **Sources.** Query those in the profile (logs, traces, database, errors) in parallel, read only. Code sweep: `tune-up:mapper` (haiku, medium unless `models` says otherwise) on today's `origin/<main>`. Check a trace/log exists before promising to find it.
4. **Timeline.** Build customer -> system, with the time of each event. Mark where the chain breaks.
5. **Classify the cause.** Code x data x configuration/prompt x external service. Say which and why; if undecidable, say what is left to measure.
6. **Hypothesis with proof.** Cite `file:line` from today's `origin/<main>` and the data that confirms it. A hypothesis without proof is marked "assumption".
7. **Smallest possible fix.** A few lines + 1 test that fails without the fix. No refactor.
8. **Phase in progress?** If a phase is being planned, ask in one line: "goes into phase NNN's plan, or fix now?"
9. **Fix only on order.** With an order: new branch from main, PR through `/tune-up:ship`. Re-enabling a conversation, resending a message and production writes need the owner's request in the turn.
10. **Memory.** Transient, external or closed cause: record it with "do NOT reinvestigate" and the evidence, so the next session does not redo it.

## Answer

- Cause in 1-2 sentences (or "not found; missing X").
- Timeline in 3-6 lines.
- Proposed minimal fix and what depends on the owner.
- Where the report is (absolute scratchpad path), if one was written.

From stage2dev/tune-up, MIT.

More from stage2dev/tune-up

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Oct 20, 2026 (8 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from stage2dev

Cite this page

investigate skill. Agent Mart. https://agentmart.dev/skills/stage2dev-investigate (last updated 2026-10-24)