The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/infer_destination.py utilizes subprocess.run to execute git commands (e.g., git rev-parse --show-toplevel) for identifying repository roots and verifying git status. This behavior is consistent with its stated purpose of managing skill placement within development environments.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and modify external skill packages, creating an attack surface where malicious instructions could be ingested from analyzed files.
Ingestion points: The skill reads SKILL.md files, evaluation JSON, and repository metadata from user-provided directories.
Boundary markers: Instructions include explicit test cases (e.g., in evals/evals.json) designed to detect and reject attempts at authority expansion or credential theft within processed data.
Capability inventory: The skill performs local file system operations (read/write) and executes validation scripts and unit tests via the unittest framework.
Sanitization: The skill advocates for grounding all generated content in authoritative sources (e.g., official documentation) and using read-only verification routes.
[EXTERNAL_DOWNLOADS]: The skill's documentation in references/cross-harness.md references the use of npx --yes skills for verifying skill discovery. This involves fetching and running a package from the public npm registry as part of a developer workflow.

skill-creator-advanced