gh-fix-ci
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads external, potentially untrusted GitHub Actions run logs and API log outputs to diagnose failures, then uses these summaries to draft and implement code changes. If an attacker controls the output printed during a CI run, they could attempt an indirect prompt injection attack to manipulate the agent's code modification workflow.
- Ingestion Points: GitHub Actions logs fetched via
gh run view <run_id> --logand the GitHub Jobs API (SKILL.md). - Boundary Markers: Absent; no explicit boundaries or delimiter instructions are defined for the parsed log outputs.
- Capability Inventory: Ability to modify local repository code and apply fixes based on the plan drafted from logs.
- Sanitization: No verification or filtering of log data content is explicitly implemented.
Audit Metadata