The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the git command-line tool via subprocess.check_output in scripts/reconcile.py to identify changed files and resolve commit references. This behavior is standard for development tools that require repository context and is implemented using safe argument passing to prevent shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill parses repository files, including tasks.md and INDEX.md, to identify new capabilities and update the feature catalog. While processing external repository content introduces a surface for indirect prompt injection, the skill mitigates this risk through robust extraction regex, strict path validation, and a mandatory confirmation step ('confirm-then-write') before applying any changes to the filesystem.
[DATA_EXPOSURE]: The skill maintains internal state in a .skills/ directory. It includes logic to verify that this directory is ignored by Git before performing writes, preventing the accidental exposure or commitment of agent-specific metadata to the repository.

map-features