The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically computes its installation path by searching the $HOME/.claude/plugins/cache/ directory. It then uses source to execute a shell script (_resolve-plugin-root.sh) and node to execute a JavaScript file (build-graph.mjs) from this computed location. This pattern involves executing code from paths determined at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill processes multiple data sources including scan-manifest.json, coverage reports, and layer documentation (docs/unwind/layers/**/*.md) to construct a knowledge graph.
Ingestion points: Files under docs/unwind/ are read and their contents (file/function/class names, priority tags, and documentation references) are used to populate the graph.
Boundary markers: The skill does not explicitly define delimiters or instructions to ignore potential commands embedded within the documentation files or progress overlays.
Capability inventory: The skill can execute Node.js scripts and write files to the docs/unwind/ directory.
Sanitization: There is no mention of sanitization or validation of the natural language instructions or tags (e.g., MUST/SHOULD/DON'T) extracted from markdown files before they are merged into the final artifact.

uw-graph