caveman-token-optimizer
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install components from an external GitHub repository (
github.com/JuliusBrussee/caveman) using various methods includingnpx skills add,git clone, andpip install. - [INDIRECT_PROMPT_INJECTION]: The skill modifies agent behavior by injecting instructions into the system prompt, which creates a surface for indirect prompt injection if the source repository content is modified.
- Ingestion points: The
SKILL.mdfile is designed to be loaded into the agent's system context as shown in the Python integration example. - Boundary markers: The provided implementation example does not utilize delimiters or specific instructions to isolate the skill content from the rest of the system prompt.
- Capability inventory: The skill documentation references capabilities including environment variable configuration, package installation, and execution of local Python scripts (
run_benchmarks.py). - Sanitization: No sanitization or schema validation is performed on the
SKILL.mdcontent before it is interpolated into the agent's system prompt.
Audit Metadata