The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with untrusted repository content (PR descriptions, comments) and loads repo-local extension files, creating an indirect injection surface. * Ingestion points: Processes PR metadata and comments fetched via tracker operations, and loads an optional repository-local extension file at .ai/skills/om-approve-merge-pr/SKILL.md. * Boundary markers: Includes explicit instructions in SKILL.md and references/agentic-setup.md to treat repository content as data rather than instructions, and to skip any extensions that attempt to relax safety or quality rules. * Capability inventory: Possesses capabilities to merge code (merge-pr), submit reviews (review-pr), and create tracker issues (create-issue). * Sanitization: Implements input validation using a whitelist regex (^[A-Za-z0-9._/-]+$) for all externally-sourced values before use in shell or path interpolation.
[PROMPT_INJECTION]: The skill contains defensive instructions designed to harden the agent against injection attacks. Specifically, references/agentic-setup.md identifies common injection patterns such as "ignore previous instructions" and directs the agent to report these as suspected injection attempts rather than complying.
[COMMAND_EXECUTION]: The skill performs GitHub and tracker operations and may execute local build or test commands. These actions are restricted to the project context and are governed by the security boundaries in references/rules.md which prohibit the exposure of secrets or access to state outside the repository.

om-approve-merge-pr