The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it reads and processes untrusted content from the repository and external worker outputs to drive its logic.
Ingestion points: Reads repository structure, documentation (docs/, plans/, architecture notes), and execution results from worker agents dispatched via cli-jaw.
Boundary markers: Utilizes ## Approved Plan and Project root headers to separate contexts during task delegation, but does not include explicit safeguards to ignore malicious instructions embedded within the ingested files.
Capability inventory: Includes file system modification (Build phase), shell command execution (cli-jaw, npx), and task dispatching to subordinate agents.
Sanitization: Attestation logic relies on form-based JSON structure validation, which checks for the presence of narratives but does not sanitize those narratives against instruction injection.
[COMMAND_EXECUTION]: Orchestrates complex workflows through the execution of cli-jaw CLI commands (e.g., orchestrate, dispatch, goal, project). It also performs build-time verification using shell-based tooling like npx tsc --noEmit to validate TypeScript project state.

dev-pabcd