The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the GitHub Dependabot and CodeQL APIs, which represents a potential surface for indirect prompt injection if alert metadata were to contain malicious instructions. However, the skill implements strong mitigations by requiring fixes to be filtered against specific high-confidence rule IDs and repository-defined policies (e.g., dependabot.yml ignore rules).
Evidence Chain:
Ingestion points: GitHub API endpoints for Dependabot and CodeQL alerts specified in SKILL.md.
Boundary markers: The skill provides explicit instructions to filter alerts to 'high-confidence fixes' and respect project-level configuration files.
Capability inventory: The skill utilizes git commit to modify the codebase, make lint and go test for verification, and go tool mage for file generation.
Sanitization: All proposed fixes undergo mandatory verification via make lint and go test ./... before being committed, ensuring that only syntactically correct and functional code is integrated.
[COMMAND_EXECUTION]: The skill executes various standard development and build tools to fulfill its purpose. These include the GitHub CLI (gh) for querying alerts, git for branch management and commits, go and pnpm for dependency management and testing, and make for linting. These commands are restricted to the local development environment and are standard for software maintenance tasks.

security-remediate