The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several build and test commands at runtime to verify code changes. Specifically, the Review Agent runs npm ci, npm test, and npm run build within isolated git worktrees (../mcpvault-triage-<id>) to validate both locally generated fixes and external contributor PRs.
[EXTERNAL_DOWNLOADS]: The skill utilizes npm ci during the fan-out and review phases, which involves downloading project dependencies from the npm registry to ensure a clean build environment.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external sources.
Ingestion points: Reads failing CI logs via gh run view, issue bodies via gh issue list, and contributor comments via gh issue view.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when passing this data to the Draft or Review sub-agents.
Capability inventory: The skill possesses write access to the repository through gh pr create, gh issue comment, and git worktree operations.
Sanitization: There is no mention of sanitizing or escaping the content of logs or issues before they are processed by the AI agents.
[REMOTE_CODE_EXECUTION]: The skill implements a 'Review' mechanism that checks out contributor PRs and executes their associated build and test scripts. This represents a risk where malicious code in an untrusted PR could be executed on the machine running the triage loop. This risk is mitigated by the skill's design, which uses isolated worktrees and requires manual human approval for any merge operations.

triage