crysk-ci-setup
Warn
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill frequently uses
npx -y crysk <command>, which downloads and executes thecryskpackage from the NPM registry at runtime. This provides a vector for executing code from an external source during the setup process. - [DATA_EXFILTRATION]: Instructions in
references/opencode-ci.mdprompt the agent to read the localauth.jsonfile at~/.local/share/opencode/auth.jsonusing a Python script. This file contains sensitive authentication tokens and provider configurations. - [PRIVILEGE_ESCALATION]: The skill performs high-privilege operations such as setting GitHub repository secrets (
gh secret set) and modifying branch protection rules via the GitHub API (gh api -X PUT ... branches/<default>/protection). - [PERSISTENCE]: The skill scaffolds a persistent GitHub Actions workflow (
.github/workflows/crysk.yml) that executes on repository events like pull requests and comments, ensuring code execution in the CI environment. - [INDIRECT_PROMPT_INJECTION]: The CI workflow reacts to pull request comments (e.g.,
/crysk review). - Ingestion points: Pull request comments (
issue_commentevent) and pull request diffs. - Boundary markers: The skill relies on GitHub's write-access checks for comment authors but lacks explicit delimiters to prevent malicious instructions within the PR content from influencing the agent's behavior.
- Capability inventory: Access to GitHub tokens, repository secrets, and status check manipulation.
- Sanitization: Permission verification for command triggers is implemented, but content-level sanitization for ingested data is not specified.
- [DYNAMIC_EXECUTION]: The skill uses a Python script passed via a shell heredoc (`python3
- <<'PY'
) to dynamically parse local credential files. It also utilizesnpx` for dynamic execution of the CLI tool. - [COMMAND_EXECUTION]: Multiple shell commands are executed using
git,gh, andpythonto inspect repository state and apply configurations. - [EXTERNAL_DOWNLOADS]: The skill downloads and references external code including the
cryskNPM package and GitHub Actions likecryskai/crysk@v0andactions/checkout@v4.
Audit Metadata