The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill orchestrates subagents by pasting data extracted from the codebase (found in docs/unwind/.cache/seeds/[layer].json) directly into their task prompts.
Ingestion points: Manifest seed files located in docs/unwind/.cache/seeds/ are used as checklists in Step 3.
Boundary markers: No delimiters or specific instructions to ignore embedded commands are used when interpolating this deterministic scan data into subagent prompts.
Capability inventory: Subagents are granted Read, Write, Edit, and Bash access to documentation directories and utilize the Task tool.
Sanitization: Content from the seed files is not sanitized or validated before being included in the prompt body.
[COMMAND_EXECUTION]: The skill's internal workflow executes local scripts to generate data and verify coverage.
Evidence: The process involves sourcing _resolve-plugin-root.sh and running seed-layers.mjs and verify-coverage.mjs with node.
The location of these scripts is determined by searching the environment variables and the user's local plugin cache directories at runtime.

uw-analyze