The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub issue bodies and comments. It implements exemplary mitigations by treating all external text as data only, using boundary markers (fencing) when presenting data to the user, and explicitly instructing the agent to ignore embedded imperatives. All mutations are staged for human approval in a separate turn.
[COMMAND_EXECUTION]: The skill executes various shell and Node.js scripts to interact with the GitHub API. It follows best practices for preventing command injection by never interpolating untrusted prose into shell commands, using file-based or stdin-based data passing instead (e.g., using --body-file or -F body=@- with the gh CLI).
[DATA_EXFILTRATION]: The skill interacts with the GitHub API via the official gh CLI. Network operations are limited to these official API interactions, and no sensitive local data (such as credentials or environment variables) is transmitted to external endpoints.

grooming-backlogs