The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on shell command execution to perform repository management and interaction with external hosting platforms. Instructions in references/mr-ops.md provide complex shell patterns involving git, gh, glab, jq, and cat. Specific capabilities include:
Managing git worktrees and branches (git worktree add, git push).
Fetching and parsing remote repository state using CLI tools.
Performing GraphQL and REST API mutations to resolve PR review threads and post comments.
Orchestrating multi-step workflows like creating draft MRs with dynamically generated titles and bodies.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to react to project feedback, making it susceptible to malicious instructions embedded in the data it processes.
Ingestion points: The skill fetches data from multiple external sources, including Git repository files, issue lists, CI status checks (gh pr checks), and PR review comments/discussions (gh api graphql).
Boundary markers: While the instructions advise the agent to "judiciously judge" (逐条判断) feedback and limit bot interaction rounds, there are no formal boundary markers or sanitization requirements for the processed text.
Capability inventory: The agent has permissions to write to the filesystem (DO.md, source code), execute shell commands, and push code changes/API updates to remote repositories.
Sanitization: The instructions do not define sanitization or escaping protocols for external content interpolated into commands or documentation, relying solely on the agent's internal logic.

do-something