trace-grader
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes a local Python script (
harness/scripts/trace-grader.py) to execute its deterministic grading engine and rule checks. - [PROMPT_INJECTION]: The skill demonstrates a surface for indirect prompt injection as it processes session transcripts and audit logs which are external, potentially untrusted inputs.
- Ingestion points: Processes data from
traces.json, audit logs in.claude/audit/, and full session transcripts in~/.claude/projects/. - Boundary markers: The instructions do not specify any delimiters or directives to ignore embedded instructions within the parsed logs or transcripts.
- Capability inventory: The skill performs file system read operations and executes shell commands to run the grading script.
- Sanitization: There is no mention of sanitization, filtering, or validation of the content extracted from the agent transcripts.
Audit Metadata