The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous shell commands for searching, validating, and managing plugins, including executing local scripts like check_marketplace.sh and check_doc_skill_lists.py. It also utilizes subprocesses in its test suite to verify documentation consistency.
[DATA_EXFILTRATION]: The skill identifies and suggests reading local data from ~/.claude/projects/ (Claude Code session transcripts) for debugging purposes. This access is governed by a policy requiring user approval before mining private history.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from SKILL.md and marketplace.json files within target repositories.
Ingestion points: Reads SKILL.md frontmatter, README.md, and marketplace.json from the repository being analyzed.
Boundary markers: None identified.
Capability inventory: Significant filesystem access and shell execution capabilities.
Sanitization: Uses standard JSON and regex parsing, but lacks specific sanitization for potentially malicious prompt content embedded in external skill descriptions.

marketplace-dev