skill-testing
Pass
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill defines procedures for executing various shell commands to perform mechanical verification of skill outputs. Evidence includes the use of
grep,ls,wc,git,cp,cmp, andpython3 -cwithinreferences/execution.mdandreferences/verification.mdfor tasks such as idempotency checking, file comparison, and arithmetic validation. - [PROMPT_INJECTION]: Indirect Prompt Injection Surface: The skill processes untrusted 'target' skills and their documentation to extract rules and inputs, which could be used to influence the testing agent's behavior.
- Ingestion points: Target skill files (
SKILL.md) and user-provided inputs as defined inSKILL.md(Step 1 and Step 3). - Boundary markers: The skill uses a 'minimal-prompt template' for the subagent, but does not explicitly define character-level delimiters or sanitization for the extracted rules.
- Capability inventory: Access to the
Agenttool (subagent dispatch), shell command execution, and local file system interaction (Step 4). - Sanitization: The skill limits verification to 'mechanical' rules (grep/regex/count/diff) which provides a narrow interpretation layer, reducing the risk of accidental instruction obedience.
Audit Metadata