quick-angle

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input and external readout files, creating a potential surface for indirect prompt injection. However, it incorporates robust mitigations:
  • Ingestion points: User input in the 'Grill' section and content from .agents/kickoff-readout.md (SKILL.md).
  • Boundary markers: The skill strictly enforces a positioning/v1 schema and uses clear section headers in Chinese to delimit data (references/angle-doc-template.md).
  • Capability inventory: File write operations to .agents/positioning.md; no network or command execution capabilities are present.
  • Sanitization: The skill explicitly mandates an 'Input hygiene' pass via ../shared/references/input-hygiene.md to remove PII, URLs, and agent instructions before content is written (SKILL.md).
  • [DATA_EXFILTRATION]: The skill features a strict 'No external research' rule, explicitly forbidding the fetching of URLs or verification of claims via network operations (SKILL.md). No network-capable tools are utilized.
  • [OBFUSCATION]: Analysis of the markdown headers and template content confirmed that Chinese character sections (e.g., '## 要打誰') are legitimate non-English content representing the project schema and do not contain homoglyphs or hidden characters (references/angle-doc-template.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 12:03 AM
Security Audit — agent-trust-hub — quick-angle