ce-polish
Warn
Audited by Gen Agent Trust Hub on Jul 12, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes development server start commands such as 'npm run dev', 'bin/dev', or 'overmind start' based on configurations found in the user's project files. These commands are executed in the local workspace as part of the 'Polish' workflow defined in 'SKILL.md'.
- [DATA_EXFILTRATION]: The script 'scripts/resolve-port.sh' reads '.env', '.env.local', and '.env.development' files at the repository root to extract the 'PORT' variable. Accessing these sensitive environment files, even for configuration purposes, is flagged as an exposure risk.
- [PROMPT_INJECTION]: The skill is subject to indirect prompt injection through untrusted project metadata and configuration files.
- Ingestion points: Project files including 'package.json', '.env', '.claude/launch.json', and 'Procfile.dev' are read by 'scripts/detect-project-type.sh', 'scripts/read-launch-json.sh', and 'scripts/resolve-port.sh'.
- Boundary markers: No boundary markers are used to isolate or ignore potentially malicious instructions embedded in these files.
- Capability inventory: Subprocess execution via 'bash' (found in scripts/resolve-port.sh and scripts/detect-project-type.sh) and web interaction via 'agent-browser' (mentioned in SKILL.md).
- Sanitization: Port extraction in 'scripts/resolve-port.sh' uses a regex for numeric values, but the skill lacks validation for command names or script arguments loaded from external project files.
Audit Metadata