Security risk report for @forgecat/openai_skills_gh-fix-ci v0.0.11
Source Integrity
Low
Profile is sourced from OpenAI's official skills repository (https://github.com/openai/skills) with a verifiable commit hash.
Metadata indicates a curated, well-maintained skill with multi-platform testing (Claude Code, Cursor, Codex).
Agent Intent
Low
Content describes a legitimate CI/CD debugging workflow with explicit approval gates ('implement only after explicit approval').
No instructions to exfiltrate credentials, read sensitive files, ignore system prompts, or install malicious dependencies.
Guidance is standard and security-conscious: authenticate first, scope to GitHub Actions only, treat external providers as out-of-scope.
Details
Evidence
"implement only after explicit approval"
"Treat external providers (for example Buildkite) as out of scope"
"Verify gh authentication" as first step in workflow
Permissions
PRM-000Medium
The skill invokes shell commands (`gh`, `python`, `gh api`) and reads log files, which constitutes shell/file-read authority; however, this is consistent with the stated CI-debugging purpose.
The `gh api` call to fetch job logs writes output to a local path (`> "<path>"`), which is a file-write operation, though scoped to log data rather than arbitrary filesystem mutation.
No tools are formally declared, so the authority is exercised implicitly through agent shell access rather than through explicitly scoped tool grants — slightly broader than ideal but not egregiously excessive for a CI debugging skill.
Details
Evidence
"gh api \"/repos/<owner>/<repo>/actions/jobs/<job_id>/logs\" > \"<path>\""
ForgeCat