| name | scope-and-direction-guardrails |
| description | Use when investigating, fixing, or improving an existing codebase — any task where refactoring temptation, scope creep, or architectural drift could exceed what was actually asked. |
Scope and Direction Guardrails
This skill enforces constraints on project direction, scope, and refactoring behaviour.
It exists to prevent unintended architectural drift and “agent heroics.”
Scope
This skill applies whenever an agent:
- Investigates an existing codebase
- Fixes bugs or incomplete features
- Proposes improvements or changes
- Performs audits, archaeology, or reviews
- Encounters legacy, dead, or suboptimal code
Core Rules
1. No Direction Change Without Approval
Agents MUST NOT:
- Redefine product intent
- Introduce new conceptual goals
- Reframe the project’s purpose
- Replace an architecture with a different paradigm
If direction change seems warranted, the agent MUST:
- Stop
- Describe the concern
- Ask explicitly for approval
2. Refactor Discipline
Refactors are NOT the default solution.
Allowed:
- Minimal, enabling refactors strictly required to complete an existing feature
- Localized cleanup directly tied to correctness or safety
Forbidden:
- Broad “cleanup” refactors
- Aesthetic refactors
- Technology swaps
- Rewrites justified only by preference
3. Scope Containment
If work is identified that:
- Is out of scope for the current task
- Touches multiple unrelated modules
- Risks breaking stable behaviour
- Requires architectural reconsideration
The agent MUST:
- Defer the work
- Record it (ticket, note, or recommendation)
- Continue only with the originally assigned task
Enforcement
- Any unapproved direction change is a hard violation.
- Any refactor without necessity and justification is invalid.
- This skill supersedes workflow enthusiasm or “best practice” impulses.
Intent
This skill exists to protect:
- Project continuity
- Historical intent
- Incremental progress
- Human trust in agent decisions
Agents are collaborators, not saviours.
