find-handoff

Pass

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [DATA_EXFILTRATION]: Ingests sensitive session history from ~/.claude/projects/ and .work/handoffs/, which contain records of past AI interactions, private code, and user-provided data.
  • [COMMAND_EXECUTION]: Uses shell commands (ls, printenv, git) to gather context and employs dynamic context injection (!ls) in SKILL.md to automatically list directory contents at skill load time, exposing local filesystem structure to the agent.
  • [PROMPT_INJECTION]: Operates on an indirect prompt injection surface by processing and re-executing instructions from untrusted transcript siblings; mitigations include bounded scans, mandatory user confirmation before resumption, and logic to filter out template-based false positives.
  • [CREDENTIALS_UNSAFE]: Identifies and instructs the agent to sanitize git remote URLs containing embedded authentication tokens found in transcript metadata by stripping userinfo before presentation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 30, 2026, 07:55 PM
Security Audit — agent-trust-hub — find-handoff