The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources, specifically pull request (PR) review comments and CI logs. * Ingestion points: references/steward-runtime.md describes fetching PR activity, including unresolved human review comments from the hosting platform. * Boundary markers: The skill uses a structured JSON schema defined in references/handoff-format.md to pass information between the Steward and the Parent agent, which acts as a data delimiter. * Capability inventory: The Parent agent has the capability to modify the codebase based on escalated items, while the Steward can execute shell commands for linting and testing. * Sanitization: The triage-rubric.md mandates that all human comments be treated as Tier C, requiring manual review and decision-making by the Parent agent or user rather than autonomous execution.
[COMMAND_EXECUTION]: The skill requires the agent to execute various shell commands to perform linting, formatting, testing, and repository management operations. Evidence includes references to tools like eslint, prettier, and npm test in references/repo-conventions.md, as well as the use of the GitHub CLI (gh) for viewing PR state in references/steward-runtime.md. These commands are standard for the skill's intended purpose in a software development environment.

pr-steward