# Stay In Scope

> Restore the active request's authorization boundary when work may be drifting or the user asks to stop scope creep.

- Skill: `tzachbon/stay-in-scope` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add tzachbon/stay-in-scope`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tzachbon/stay-in-scope/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: tzachbon (https://skillmd.com/u/tzachbon)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/tzachbon/stay-in-scope

---


# Stay in scope

Use this as a manual recovery command when work may be drifting.

1. State the active scope envelope in one compact line: target, action, bounds, deliverable, and stop condition.
2. Use the narrowest clear reading. An explicit broad request remains broad.
3. Identify any work already started outside the envelope. Stop that work and preserve its state when doing so avoids data loss.
4. Return to the requested task and finish the authorized work that remains.
5. Preserve verb boundaries: draft does not authorize sending, review or diagnosis does not authorize fixes, planning does not authorize execution, and one artifact does not authorize changes to its collection.
6. Read broadly when context requires it. Keep writes, external actions, and side effects inside the envelope.
7. Resolve facts with read-only inspection. Ask one concise question before mutation only when two plausible readings would change the envelope.

Invoking this skill does not authorize a mutation, send, fix, implementation, or expansion by itself.