invisible-ink

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONOBFUSCATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted repository files, which constitutes an attack surface for indirect prompt injection. The skill implements strong mitigations by redacting potential secrets (passwords, tokens), escaping all invisible or control characters in its output, and providing explicit instructions to the agent to treat decoded content as untrusted data rather than instructions.
  • [PROMPT_INJECTION]: The documentation in SKILL.md contains phrases typical of prompt injection attacks, such as "ignore previous instructions." These instances are presented within example output blocks to illustrate the tool's detection capabilities and do not represent an attempt to override the agent's core instructions.
  • [OBFUSCATION]: The scanner script scripts/invisible_ink.py contains definitions for various invisible and deceptive Unicode character ranges, including Trojan Source bidi controls and homoglyph scripts. These are used strictly for detection logic and are represented using numeric ranges or character escapes rather than raw obfuscated payloads.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 12:23 PM