The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub Action logs, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The skill uses gh run view <run-id> --log-failed in SKILL.md to fetch logs from potentially attacker-controlled CI environments.
Boundary markers: There are no explicit delimiters or specific instructions for the agent to disregard embedded content within the logs.
Capability inventory: The skill has authority to execute GitHub CLI commands, including repository search, configuration changes, and reversible writes.
Sanitization: The workflow uses rg (ripgrep) to filter for specific patterns like error, FAIL, or panic. This limits the amount of raw log data processed but does not eliminate the risk of a crafted injection payload appearing within those filtered lines.
[COMMAND_EXECUTION]: The skill is primarily designed to interact with the system by executing gh CLI commands.
It uses various subcommands like run, search, auth, config, and issue to perform its tasks.
Safety instructions are included to limit authority (e.g., 'Never delete repositories') and require approval for label deletions.

cli-gh