systematic-debugging
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data such as error logs and test output, which could contain instructions to manipulate the agent. 1. Ingestion points: SKILL.md instructs the agent to capture and read full error messages. 2. Boundary markers: No explicit instructions are provided to delimit or ignore potential commands embedded within the analyzed logs or error output. 3. Capability inventory: The agent is instructed to use shell-based tools including git (log, diff, show, revert) and package managers (npm install, npm test). 4. Sanitization: There is no evidence of input validation or content filtering for the ingested logs before processing.
- [COMMAND_EXECUTION]: The skill workflow involves running local commands to analyze code state and test fixes. Evidence: references/root-cause-tracing.md and references/defense-in-depth.md provide instructions for using git and npm tools for history analysis and environment isolation.
Audit Metadata