# Investigation Modes

> Chooses and sticks to an investigation mode (debug, explore, compare) with matching depth and output shape. Use this skill when starting analysis of a bug, unfamiliar area, or trade-off and mode selection would otherwise thrash. Do not use when/for implementing a fix already diagnosed, or session handoff notes alone (use session-continuity).

- Skill: `jagreehal/investigation-modes` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add jagreehal/investigation-modes`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jagreehal/investigation-modes/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: jagreehal (https://skillmd.com/u/jagreehal)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/jagreehal/investigation-modes

---


# Investigation Modes

## Critical rules

- Operate in one mode at a time: LEARNING | INVESTIGATION | SOLVING.
- Prefix every message: `[MODE: LEARNING]`, `[MODE: INVESTIGATION]`, or `[MODE: SOLVING]`.
- Ask before transitioning. User controls pace: "do X, THEN Y" = finish X, STOP, ask before Y.
- Stay in lane: no hypotheses/fixes in LEARNING; no edits in INVESTIGATION.
- Before applying mode boundaries or examples, read [references/modes.md](references/modes.md).

## Workflow

1. If mode unclear, ask the user to choose LEARNING / INVESTIGATION / SOLVING; wait.
2. Prefix every message with the active mode.
3. Produce only that mode's outputs; respect boundaries (see modes reference).
4. On completion of a phase, ask before moving to the next mode.
5. If user said THEN: stop after the first phase and wait.

## Resources

- [references/modes.md](references/modes.md) — mode definitions, state machine, examples, RCA/USE notes. Read when selecting or switching modes.

## Validation

- [ ] Message prefixed with current mode
- [ ] Within mode boundaries
- [ ] User authorized this mode
- [ ] Transitions asked first
- [ ] "THEN" respected (stopped after first part)

## Constraints

- Skip for trivial changes where cause and fix are already obvious.
- Related: `debugging-methodology`, `design-exploration`, `implementation-planning`, `verification-before-completion`, `session-continuity`, `confidence-levels`, `observability`.