dsh-trim-cot-leakage

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes repository prose (READMEs, JSDoc, Agent Notes) as untrusted input, creating an attack surface where malicious instructions could be embedded in the audited text.
  • Ingestion points: Prose files throughout the repository (excluding vendor/ and archived notes).
  • Boundary markers: Absent; the skill does not specify delimiters or instructions to ignore embedded commands in the source text.
  • Capability inventory: Includes reading files, executing rg (ripgrep) search commands, and writing modifications to files.
  • Sanitization: Absent; the skill relies on the agent's semantic judgment without pre-processing or escaping inputs.
  • [COMMAND_EXECUTION]: The skill requires the execution of shell commands and project-specific scripts to identify and verify changes.
  • Evidence: references/recall-batteries.md contains multiple rg (ripgrep) command templates. SKILL.md references running project "gates" such as doc-sync, verify-type-equiv, and verify-translation-pairing.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 01:53 AM
Security Audit — agent-trust-hub — dsh-trim-cot-leakage