github-pr-workflow

Warn

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONINGDYNAMIC_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill programmatically reads sensitive credential files including ~/.hermes/.env and ~/.git-credentials to extract a GitHub authentication token.
  • [INDIRECT_PROMPT_INJECTION]: The skill implements an automated CI failure fixing loop that ingests and acts upon data from external GitHub Actions logs. This untrusted data source represents a vulnerability surface where an attacker could provide malicious log output to influence the agent's code modifications or API actions.
  • Ingestion points: SKILL.md (automated log download and extraction).
  • Boundary markers: None present to delimit untrusted log data.
  • Capability inventory: write_file, patch, git commit, git push, and GitHub API write operations.
  • Sanitization: No sanitization or validation is performed on the ingested log content.
  • [METADATA_POISONING]: The author field in the skill metadata is listed as "Hermes Agent", which conflicts with the provided skill author information.
  • [DYNAMIC_EXECUTION]: The skill executes Python code snippets at runtime via python3 -c to parse JSON data retrieved from remote GitHub API endpoints.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 8, 2026, 07:01 PM
Security Audit — agent-trust-hub — github-pr-workflow