The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform environment discovery and execute internal logic. It specifically runs shell commands to find its installation path and execute helper scripts.
Evidence: The script in 'Phase A' uses ls, source, and node to locate and execute plan-brief.mjs and _resolve-plugin-root.sh.
[DYNAMIC_EXECUTION]: The skill dynamically calculates the path to its own executable scripts at runtime based on the environment's plugin cache structure.
Evidence: UNWIND_PLUGIN_ROOT="$(ls -dt "$HOME"/.claude/plugins/cache/*/unwind/*/ 2>/dev/null | head -1)" is used to resolve the script location before execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data from project manifests and local documentation to influence the agent's behavior during the user interview process.
Ingestion points: Reads external project files such as package.json, requirements.txt, go.mod, and internal cache files like plan-brief.json.
Boundary markers: None explicitly defined to separate external data from internal reasoning.
Capability inventory: The skill has permissions to write and edit documentation files (Write, Edit) and execute shell commands (Bash).
Sanitization: No explicit sanitization of values derived from project manifest files is described.

uw-plan