The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from several sources to drive its logic and subagent dispatches. \n
Ingestion points: Reads project files such as requirements.md, INDEX.md, and tasks.md in SKILL.md, as well as issue tracker data via tracker-sync.md. \n
Boundary markers: The subagent templates in implementer-prompt.md and task-reviewer-prompt.md use bracketed placeholders for external content but lack strict delimiters or safety warnings for the LLM to ignore instructions within that data. \n
Capability inventory: The skill performs shell operations using git (including rev-parse, log, diff, and worktree in task-lifecycle.md) and npx (in SKILL.md), manages local files, and interacts with external tracker APIs. \n
Sanitization: No explicit sanitization or content validation is performed on the data read from external project files before processing. \n- [COMMAND_EXECUTION]: The skill performs automated shell operations for repository and state management. \n
Evidence: Uses git commands for worktree creation and status tracking in task-lifecycle.md. It also programmatically updates the .gitignore file and commits changes in ledger-check.md.

execute-common