The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill is configured to execute local scripts and build tools as part of the PR checklist.
Evidence: Use of npm run lint, npm run validate, and pwsh -File tools/precommit_safety_scan.ps1.
Context: These are standard maintainer actions, but they execute code based on the repository's current state.
[DATA_EXFILTRATION] (SAFE): The skill includes specific defensive patterns to prevent data exposure.
Evidence: Grep patterns for C:\Users, /Users/, and /home/ to detect PII in staged files before committing.
[REMOTE_CODE_EXECUTION] (SAFE): No remote script downloads or piped executions (e.g., curl | bash) were detected.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes external data (PR content, file diffs) and has access to execution tools.
Ingestion points: git diff --cached reads staged file content into the agent context.
Boundary markers: Absent; there are no specific delimiters used when the agent processes file changes.
Capability inventory: High; includes git, npm, python, and powershell.
Sanitization: Absent; the skill relies on the user/agent to manually check for PII but does not sanitize the inputs to the tools themselves.

maintainer