harness-creator
Pass
Audited by Gen Agent Trust Hub on Jul 10, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The script
create-harness.mjsapplies executable permissions (chmod 0o755) to the generatedinit.shfile. This is a standard and necessary operation for the generated verification script to be runnable by the agent or user. - [COMMAND_EXECUTION]: The
run-benchmark.mjsscript utilizesexecFileto execute its owncreate-harness.mjsscript within a temporary directory as part of an end-to-end self-check routine. - [DATA_EXFILTRATION]: The skill reads project configuration files (e.g.,
package.json,go.mod,Cargo.toml) and specific harness files from the local filesystem to perform auditing and stack detection. No evidence of data transmission to external domains was found. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository when performing harness audits. This constitutes a potential surface for indirect prompt injection, although the risk is limited by the structural nature of the analysis and the implementation of HTML escaping in the report generation.
- Ingestion points: Reads files such as
package.json,AGENTS.md, andfeature_list.jsonfrom the target project directory. - Boundary markers: None identified in the processing logic.
- Capability inventory: Includes
writeFile,chmod, andexecFile(restricted to self-testing). - Sanitization: Uses HTML entity escaping when rendering assessment reports to prevent potential XSS from malicious file content.
Audit Metadata