explanation-traces

Pass

Audited by Gen Agent Trust Hub on Apr 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the Bash tool to run diagnostic and validation commands. Reference files such as "references/error-handling.md" and "references/preferred-patterns.md" provide snippets using "python3 -c", "grep", "find", and "rg" to check the integrity of JSON files and search for hook scripts. While these are local operations, they provide the agent with pre-written shell commands that interact with the file system.
  • [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) because it reads and displays data from "session-trace.json", which is generated by other system components (hooks). If a hook were compromised or designed to include instructions in the trace's "evidence" or "context" fields, the agent might inadvertently execute or follow those instructions during the presentation phase.
  • Ingestion points: Accesses "session-trace.json" and ".claude/session-trace.json" via the Read tool in Phase 1 and Phase 2.
  • Boundary markers: None identified. The instructions do not direct the agent to wrap the ingested trace data in protective delimiters or "ignore previous instructions" markers during Phase 3 (Presentation).
  • Capability inventory: The skill utilizes Bash, Read, Glob, and Grep tools, providing a surface for command execution if manipulated.
  • Sanitization: There is no logic provided to sanitize or filter potential instructional text from the decision trace before it is formatted and presented to the user.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 30, 2026, 12:34 PM
Security Audit — agent-trust-hub — explanation-traces