The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a dedicated security script (detect_prompt_injection.py) that proactively scans issue titles, bodies, and comments for malicious injection patterns using a scoring system before any analysis occurs.
[DATA_EXFILTRATION]: No evidence of data exfiltration was found. The skill uses environment variables for Linear credentials (LINEAR_CLIENT_ID, LINEAR_CLIENT_SECRET) within a Python script to prevent accidental exposure in logs or CLI arguments. Network operations are restricted to known official APIs (GitHub and Linear).
[COMMAND_EXECUTION]: The skill executes local utility scripts and standard tools like gh and grep. It includes specific instructions to strip shell metacharacters from issue-derived search terms in Step 3 to prevent command injection.
[EXTERNAL_DOWNLOADS]: All external references are to well-known technology platforms (GitHub API, Linear API) or the vendor's own repositories (getsentry/sentry-javascript, getsentry/sentry-docs). These are documented as safe vendor operations.
[INDIRECT_PROMPT_INJECTION]: The skill explicitly identifies issue content as untrusted data. It mitigates indirect injection risks by running a mandatory pre-check script and instructing the agent to treat issue content solely as data for classification rather than instructions to follow.

triage-issue