The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes uv pip install -e and uv run python to verify that local packages are correctly linked and can be imported. These operations are performed on the local file system within the user's project directories.
[INDIRECT_PROMPT_INJECTION]: The skill reads from pyproject.toml, CLAUDE.md, and Atlas topic files to extract names and descriptions for documentation updates.
Ingestion points: Reads pyproject.toml, CLAUDE.md, README.md, and files in ~/vault/atlas/ (SKILL.md).
Boundary markers: None present; data is directly interpolated into templates.
Capability inventory: Uses Write, Edit, and Bash(uv*) for file modification and local installation.
Sanitization: No specific sanitization or validation of the ingested metadata is performed before interpolation.
[DATA_EXPOSURE]: The skill searches for and modifies files located in ~/vault/atlas/ to maintain a record of shared infrastructure. This behavior is consistent with the skill's stated purpose of tracking dependencies across a research vault.

wire-shared-package