scbe-claude-crosstalk-workflow

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on the execution of local PowerShell (emit_crosstalk_packet.ps1) and Python (crosstalk_skill_audit.py) scripts. These scripts are invoked via absolute paths (e.g., C:/Users/issda/.codex/skills/scbe-claude-crosstalk-workflow/scripts/) that are specific to the author's environment to manage session-aware packet data.\n- [INDIRECT_PROMPT_INJECTION]: The skill facilitates inter-agent communication via shared files, which presents a surface for indirect prompt injection.\n
  • Ingestion points: Agents ingest data from communication packets stored in artifacts/agent_comm/ and notes/_inbox.md.\n
  • Boundary markers: The skill instructions do not provide delimiters or security instructions to isolate data in packet fields from the recipient agent's logic.\n
  • Capability inventory: The skill uses scripts to perform file system operations (write and audit) across the repository root.\n
  • Sanitization: There is no evidence of sanitization, validation, or escaping of the content being written to packet fields such as Summary, NextAction, or Why before it is consumed by other agents.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 07:01 PM
Security Audit — agent-trust-hub — scbe-claude-crosstalk-workflow