# Bounded Investigation

> Use when debugging, architecture inspection, research, or repository exploration risks becoming open-ended, recursive, speculative, or disconnected from a decision.

- Skill: `ihabkhaled/bounded-investigation` (Agent Skill)
- Install (CLI): `npx skillmds@latest add ihabkhaled/bounded-investigation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ihabkhaled/bounded-investigation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: ihabkhaled (https://skillmd.com/u/ihabkhaled)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/ihabkhaled/bounded-investigation

---


# Bounded Investigation

## Core principle

An investigation exists to answer one decision-enabling question, not to understand everything.

## Procedure

1. State the exact question.
2. Name the evidence source and expected decision.
3. Set maximum depth, time or attempt budget, and stop condition.
4. Search first, then read the smallest relevant source.
5. Record what the evidence proved or disproved.
6. Return the decision to the parent task; park anything else.

## Required output

Question, evidence, conclusion, parent impact, and next recommendation.

## Limits

Nesting at most 3; same strategy at most 3 attempts.

## Common mistakes

Do not replace the question with a broad audit or preserve a failed hypothesis because of sunk cost.

## Stop condition

Stop when the decision is possible, the budget is reached, or required evidence is externally unavailable.