nuke-docs
Fail
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill mandates that documentation claims must be verified by running them directly. Mandate 1 in SKILL.md states that a command claim is verified by running it or checking its help output. This allows the execution of arbitrary shell commands found in documentation files.
- [REMOTE_CODE_EXECUTION]: The skill compiles and executes code snippets found in documentation. In SKILL.md (Full mode) and Mandate 1, API examples are verified by compiling and running them, which allows for complex runtime code execution.
- [EXTERNAL_DOWNLOADS]: During its preflight phase, the skill is permitted to automatically install dependencies. references/preflight.md specifies that environmental blockers, such as missing tooling, should be resolved by fixing the environment, including installing dependencies.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by design. 1. Ingestion points: The repository's documentation files (e.g., README, docs/ directory). 2. Boundary markers: None; claims are extracted verbatim. 3. Capability inventory: Shell command execution, code compilation, and dependency installation. 4. Sanitization: The skill relies on basic heuristics to identify destructive commands, which can be bypassed.
Recommendations
- AI detected serious security threats
Audit Metadata