create-ol-pull-request
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from commit histories and user-provided descriptions to populate a pull request template. This content is subsequently used by the agent to perform factual audits using external monitoring tools, creating a surface where malicious instructions in a commit message could influence the agent's verification actions.
- Ingestion points: Untrusted data enters the context through
git logoutput and user prompts in Step 3 (SKILL.md). - Boundary markers: The skill lacks explicit delimiters or instructions to treat commit data as untrusted when the agent performs the 'Audit factual claims' step.
- Capability inventory: The skill uses shell commands (
git,gh) and accesses monitoring infrastructure (Prometheus/Grafana viatoolhive-swe-*MCP tools) as documented in SKILL.md. - Sanitization: While the skill instructs to strip HTML comments, it does not provide mechanisms for escaping shell-sensitive characters or validating data before interpolation into the PR body or audit queries.
- [COMMAND_EXECUTION]: The skill uses
gitand the GitHub CLI (gh) to inspect branch status, check for existing PRs, and create new pull requests. These operations are standard for development workflows and are restricted to the local repository context. - [EXTERNAL_DOWNLOADS]: The skill references a pull request template hosted on the mitodl organization's official GitHub repository. This is a legitimate reference to a trusted vendor resource consistent with the skill's primary purpose.
Audit Metadata