The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for agents to automatically triage issues and review pull requests. These agents ingest untrusted content from external contributors (issue descriptions, PR comments, and code diffs). An attacker could embed malicious instructions within an issue to manipulate the agent's behavior, such as mis-prioritizing tasks, closing legitimate bugs, or influencing project board placement.
Ingestion points: triage-agent analyzes issue titles and bodies; code-reviewer analyzes pull request content and metadata.
Boundary markers: The provided prompt templates do not include explicit delimiters (e.g., XML tags) or instructions to treat input as data rather than instructions.
Capability inventory: The agents are granted github-issue_write, github-projects_write, and pull-requests: write permissions, allowing them to modify repository state based on processed data.
Sanitization: No evidence of sanitization, filtering, or validation of external content before it is passed to the LLM context.
[SAFE]: The documentation follows security best practices for CI/CD, such as recommending the use of step-security/harden-runner for egress traffic auditing and using standard GitHub Actions secrets (secrets.GITHUB_TOKEN) for authentication.
[SAFE]: The external URLs referenced in the code snippets (e.g., feedback-api.example.com) and the tools mentioned (e.g., jscpd, complexity-report) are standard documentation placeholders or well-known development utilities.

GitHub Agentic Workflows Continuous AI Patterns