The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes implementation plans (phase-N.md) and project source files which could contain indirect prompt injections targeting the orchestrator or its sub-agents. \n- Ingestion points: Implementation plan files (phase-N.md), project README, and source files modified during tasks. \n- Boundary markers: The skill uses structured markdown templates (as seen in reference/implement-flow.md Section 15) to separate task briefs from project context. \n- Capability inventory: The orchestrator utilizes Bash, PowerShell, Write, Edit, and the Agent tool to dispatch sub-agents with broad permissions. \n- Sanitization: The skill does not explicitly sanitize the contents of the plan files before passing them to agents, relying instead on structural delimiters. \n- [COMMAND_EXECUTION]: The orchestrator automatically detects and executes shell commands for typechecking, building, testing, and linting based on project configuration files (e.g., package.json, Cargo.toml). \n- Evidence: Steps 7 and 8 in SKILL.md and Section 8 in reference/implement-flow.md describe the detection of package managers and the automated execution of associated CLI tools via Bash or PowerShell.

implement