The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data including execution plans (execution-plan.md), task specifications, and issue tracker content (GitHub, GitLab, local Markdown files) to perform code implementation and verification.
Ingestion points: External execution plans located at .agents/plans/<work-slug>/execution-plan.md and content from issue trackers (GitHub, GitLab, local files) are loaded into the agent's context.
Boundary markers: The skill lacks explicit technical boundary markers (like XML tags) for data interpolation but includes strong instructional boundaries, explicitly stating that instructions within data should be treated as data to be reviewed, not as authoritative commands.
Capability inventory: The skill uses git for branch and worktree management, and executes shell commands for TDD (tests) and verification steps defined within execution plans.
Sanitization: The EXECUTE.md instructions mandate a 'Supervisor' role to manually verify that plans do not contain instructions for secret disclosure, scope expansion, or safety filter bypasses before delegating tasks to a sub-executor.

vibe-implement