investigatelisted
Install: claude install-skill bwkw/dotagents
# /investigate — answer the question, then stop
Two failure modes bracket codebase investigation. One is stopping too early and answering from a
plausible guess. The other is "look into the codebase" with no boundary, reading three hundred files
and filling the context before any work starts.
This skill fixes both ends: a **declared scope contract** before reading anything, and a **numeric
budget** that ends the search whether or not the answer is complete.
**This skill never modifies anything.** It reads and reports.
## Preconditions
| Condition | If unmet |
|---|---|
| There is a specific question to answer | **Stop.** "Investigate the auth module" is not a question. Ask what decision the answer serves — that determines when to stop. |
## Position in the workflow
| Upstream | This skill | Downstream |
|---|---|---|
| a question, or the start of planning | `/investigate` | `/grill-me`, `/writing-plans`, `/design-review` |
## Files to read
### Always read
| File | Why |
|---|---|
| `${CLAUDE_SKILL_DIR}/reference/evidence-rules.md` | How to report what you found, and what you did not |
### Read only if
| File | Trigger condition |
|---|---|
| `CLAUDE.md`, `AGENTS.md` | When the question concerns project convention rather than mechanism |
---
## Step 1. Declare the scope contract — before reading anything
State, and show the user:
- **The question**, restated precisely.
- **The decision it serves.** This is what tells you when you have enough. "Where is tenant f