The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skills ingest untrusted input from users and repository content to perform tasks like lint rule generation.
Ingestion points: The $ARGUMENTS placeholder in SKILL.md and repository files read by the Read tool.
Boundary markers: Structured output contracts using vigiles:ok and vigiles:err blocks are used to isolate results.
Capability inventory: Access to Bash, Write, and Edit tools for project maintenance.
Sanitization: The project uses a subagent railway pattern to validate implementation steps and normalize shell commands for safety checks.
[COMMAND_EXECUTION]: Automated tasks such as linting and testing are performed via the Bash tool.
Evidence: ship-pr/SKILL.md executes npm run lint and npm test to verify code quality before opening a pull request.
[EXTERNAL_DOWNLOADS]: The testing harness utilizes well-known Docker images for isolated database testing.
Evidence: harness/measure-with-service.mjs pulls the postgres:16-alpine image to verify database migrations.
[DATA_EXFILTRATION]: Egress monitoring is implemented to track and block unauthorized network requests.
Evidence: harness/oh-my-claudecode-egress.harness.mjs records update checks to registry.npmjs.org, which is documented as a neutral finding for a well-known service.

pr-to-lint-rule