article-refactor

Pass

Audited by Gen Agent Trust Hub on May 12, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted external data from user-provided drafts or articles, which creates a potential surface for indirect prompt injection.
  • Ingestion points: Files read from paths specified in the <source-path> argument or provided interactively during the extraction phase.
  • Boundary markers: The instructions define a structured 'Inventory template' and mandate a transformation process that separates source content from the agent's task logic.
  • Capability inventory: The skill performs local file system read operations for source material and write operations for generating metadata files (INVENTORY.md, SKELETON.md) and draft revisions.
  • Sanitization: The methodology explicitly instructs the agent to record information units rather than prose expression and to strip away the source's rhetoric, acting as a semantic filter against embedded instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
May 12, 2026, 01:36 PM
Security Audit — agent-trust-hub — article-refactor