The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to triage and act upon untrusted data from automated PR review comments provided by various external services such as gemini-code-assist, Copilot, and Sonar.\n
Ingestion points: Automated review comments from third-party bots are processed as input for decision-making in standards/pr-comment-disposition.md.\n
Boundary markers: The instructions state that all comments must first pass a "validity check from persona-plan-marshall-agent" and be verified against the plan's intent before any disposition is made.\n
Capability inventory: The triage process can lead to the FIX outcome, which involves modifying the codebase and committing changes. The skill also guides the execution of linting tools that modify files.\n
Sanitization: The skill mandates the use of a "PR review hard rule" to validate suggestions against the driving project plan and existing architectural lessons to prevent the adoption of malicious or erroneous suggestions.\n- [COMMAND_EXECUTION]: The skill includes instructions for running standard JavaScript ecosystem tools for automated remediation.\n
Evidence: standards/severity.md lists commands such as npx eslint --fix . and npx prettier --write . to fix identified issues. These are standard development practices and are used within their intended functional scope.

ext-triage-js