The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several system commands for dependency management and verification. Evidence includes:
Phase 2: pip-audit --json, poetry audit, and pipenv check.
Phase 3 & 4: pip list --outdated, pip install --upgrade <package>, poetry update, and pipenv update.
Phase 5: python -c "import <package>" and pytest for installation verification. These commands can execute arbitrary code if the packages being installed or tested contain malicious scripts.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it actively ingests data from untrusted external sources to guide its actions.
Ingestion points: Uses WebSearch, mcp__Ref__ref_read_url, and mcp__context7__query-docs to find migration guides and fixes for breaking changes.
Boundary markers: The documentation does not specify the use of delimiters or 'ignore instructions' blocks when processing these search results.
Capability inventory: The agent can perform file writes (updating requirements/toml), package installations (pip install), and code execution (pytest, python -c).
Sanitization: There is no mention of sanitizing or validating the instructions found in external migration guides before application.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of external code via package managers (pip, poetry, pipenv) targeting standard registries like PyPI. Additionally, diagram.html loads the Mermaid.js library from https://cdn.jsdelivr.net/npm/mermaid/dist/mermaid.min.js, which is a well-known and trusted CDN service.

ln-713-pip-upgrader