avoid-ai-writing

Pass

Audited by Gen Agent Trust Hub on Jun 30, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is comprised solely of documentation and instructional markdown files. It contains no executable scripts (Python, Node.js, Shell), binaries, or configuration files that trigger external actions.\n- [COMMAND_EXECUTION]: There are no commands or subprocess calls defined in the skill. It relies entirely on the agent's natural language understanding to apply the provided writing guidelines.\n- [DATA_EXFILTRATION]: The skill does not perform any network operations. No URLs are called during the audit or rewrite process, and there is no logic to send user data to external servers.\n- [CREDENTIALS_UNSAFE]: No hardcoded API keys, tokens, or sensitive credentials were found in any of the files.\n- [PROMPT_INJECTION]: The instructions are scoped to writing style and editing. They do not contain 'jailbreak' patterns, instructions to ignore safety filters, or attempts to extract the agent's system prompt.\n- [EXTERNAL_DOWNLOADS]: The skill does not download external packages or remote scripts. Dependencies mentioned in the documentation (like Pangram research) are cited as intellectual sources, not technical requirements.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user text. However, because it lacks dangerous capabilities, it is classified as safe.\n
  • Ingestion points: User-supplied text to be audited (SKILL.md)\n
  • Boundary markers: Absent\n
  • Capability inventory: None (no subprocess calls, no network access, no file system writes)\n
  • Sanitization: None
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 30, 2026, 05:03 AM
Security Audit — agent-trust-hub — avoid-ai-writing