phaser-game-director
Fail
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill includes a dedicated script,
scripts/probe_asset_credentials.sh, designed to detect the presence of sensitive API keys (TRIPO_API_KEY, GEMINI_API_KEY, ELEVENLABS_API_KEY) and any variables matching the*_IMAGEGEN_MODELpattern. - [DATA_EXFILTRATION]: The
scripts/probe_asset_credentials.shscript accesses sensitive file paths including~/.env,~/.zprofile,~/.zshrc,~/.bash_profile, and~/.bashrcto identify present credentials. In an "assume-malicious" posture, accessing these files to identify present credentials is considered sensitive data exposure. - [COMMAND_EXECUTION]: The skill instructions and scripts (e.g.,
references/asset-recovery.md) promote sourcing user shell profiles into subshells to execute commands. This pattern can lead to the execution of any arbitrary code stored in those user profiles. - [INDIRECT_PROMPT_INJECTION]: The
scripts/check_evidence.pytool ingests and parses external data from markdown reports and JSON manifests. While it primarily checks for file existence, it lacks robust sanitization for the content it processes, creating a surface for indirect injection via manipulated project artifacts. This finding is categorized as low severity per the required evidence chain (ingestion points: reports/manifests; capability: shell execution).
Recommendations
- AI detected serious security threats
Audit Metadata