The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts and the GitHub CLI (gh) to perform its tasks. Technical review of scripts/_skill_protocol_runtime.py and scripts/publish_github_issues.py shows these executions use subprocess.run with shell=False, which effectively mitigates shell injection risks. Command arguments, including repository URLs and file paths, are validated using strict regular expressions and resolution logic.\n- [PROMPT_INJECTION]: The skill instructions include a protective directive to treat the input handoff file strictly as data and not as instructions, establishing a clear security boundary for the model. The skill also handles untrusted external data from an audit handoff file via an indirect prompt injection surface; this is mitigated through a mandatory evidence chain: integrity is verified via a SHA256 receipt check, and the workflow requires a human-in-the-loop phase where the user must review a preview.json artifact and provide an explicit publish_confirmation=yes before any issues are created on GitHub.\n- [DATA_EXFILTRATION]: The skill publishes audit findings to GitHub repositories using the gh CLI. As GitHub is a well-known service and the destination repository is explicitly provided by the user, this behavior is consistent with the skill's primary purpose and is protected by the mandatory preview and confirmation sequence.

raise-issue