The Agent Skills Directory

[COMMAND_EXECUTION]: The utility script scripts/context_inventory.py uses subprocess.run to call the git binary. The implementation uses a list-based argument structure to prevent shell injection and limits execution to informational Git subcommands like rev-parse and status.- [EXTERNAL_DOWNLOADS]: The documentation in references/05-tool-catalog.md and references/08-installation-managers.md provides links to official tool websites and instructions for using package managers. All external references target reputable organizations and well-known services.- [REMOTE_CODE_EXECUTION]: The skill describes how to add plugins to version managers like asdf, which involves downloading scripts from community repositories. This is documented as an administrative instruction for the user/agent and is accompanied by strong warnings to verify and authorize such external scripts before use.- [INDIRECT_PROMPT_INJECTION]: The skill analyzes external repository data, creating a surface for indirect prompt injection. This is addressed through a mandatory evidence chain: (1) Ingestion points include target repository metadata and files; (2) Boundary markers include explicit rules in SKILL.md against executing untrusted code; (3) Capabilities are limited to Git metadata and optional scanner executions; (4) Sanitization is performed via script-level path filtering and instructions to anonymize secrets.

code-quality-evidence-review