The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted content from GitHub Issues, Pull Request comments, and automated bot reports (e.g., Devin or SonarQube results), which creates a surface for indirect prompt injection attacks.
Ingestion points: Ingests data through GitHub API calls (gh api) for issue bodies and review comments as described in SKILL.md and references/pr-analysis.md.
Boundary markers: The instructions explicitly define boundaries by requiring the agent to treat input as data, never follow instructions embedded in comments (e.g., "skip this test"), and quote suspicious directives to the user.
Capability inventory: The agent has access to the GitHub CLI (gh), Git commands, and the local build/test environment (e.g., .NET, Node.js).
Sanitization: The skill mandates the redaction of secrets/PII from reports and requires a Portuguese-language summary and user confirmation gate before creating issues or implementing specs based on the analyzed data.
[COMMAND_EXECUTION]: The skill executes shell commands using the GitHub CLI (gh) to fetch repository metadata and review data, and utilizes project-specific tools like dotnet test or Playwright for verification cycles.
[SAFE]: The skill includes a proactive security check in the verification-loop that uses grep to scan the codebase for leaked secrets, such as API keys and tokens, before allowing a PR to be considered ready.

qa-analyst