From agent-guardrails
Validates file modifications stay within authorized scope by checking user requests, task descriptions, and dependencies. Prevents scope creep and unauthorized edits.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-guardrails:scope-validatorThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Verify you are authorized to touch the files you are about to modify.
Verify you are authorized to touch the files you are about to modify.
Only touch files within the authorized scope.
Scope is determined by:
Before ANY file modification:
NEVER do these without explicit user authorization:
.env, package.json, go.mod, etc. unless explicitly askedWhen a change requires modifying dependencies:
Recognize when scope is expanding:
When you notice scope creep: STOP and ask user for confirmation.
SCOPE CHECK: This task involves files beyond the original request.
Original scope: <files from user request>
Additional files needed: <files you want to touch>
Reason: <why these are needed>
Should I proceed with these additional files, or keep changes limited to the original scope?
When running in pi, scope is enforced by the @architectit/pi-guardrails extension:
guardrail_set_scope defines authorized file paths at session startguardrail_check_scope verifies a path is in scope before operationstool_call handlerguardrail_mcp with action validate_scope for server-side checksSee [[guardrails-core]] for the full enforcement coverage map.
Validate that the proposed file modifications are within authorized scope. Check the user request, task description, and dependency impact before allowing changes. If scope creep is detected, halt and ask for user confirmation.
docs/workflows/AGENT_EXECUTION.md — Execution protocols and scope rulesskills/four-laws/SKILL.md — Law 2: Stay in Scopenpx claudepluginhub thearchitectit/agent-guardrails-templateMandatory safety laws for AI coding agents: read before editing, stay in scope, verify before committing, and halt when uncertain. Enforces safe and reliable code modifications.
Implements scoped code changes safely: confirms scope, edits minimal relevant files, runs checks/tests, verifies, and summarizes updates/next steps.