The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes implementation plans and specifications which may originate from untrusted external sources, creating an attack surface for indirect prompt injection.
Ingestion points: SKILL.md (Plan Sources section) mentions reading local plan artifacts, requirements specifications, and chat summaries.
Boundary markers: The skill includes an "Effect And Write Boundaries" section and "Effect class" declarations to restrict agent actions to specific scopes.
Capability inventory: Shell commands including sed -i, tee, mv, cp, rm, and git checkout -- are authorized for implementation slices, and the skill may write to filesystem paths such as docs/decisions/ and docs/reports/findings/.
Sanitization: Instructions include multiple verification gates ("Plan Validity Gate", "Post-Implementation Review Gate") and a "Startup Consent Preflight" to confirm actions and verify evidence before execution.
[COMMAND_EXECUTION]: The skill explicitly authorizes the use of local shell commands for file manipulation and state-changing operations within defined implementation boundaries.
Evidence: SKILL.md lists sed -i, tee, mv, cp, rm, and git checkout -- as permitted tools for executing plan slices.

vibe-plan-execution