improve-animations

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill contains specific countermeasures against prompt injection. Rule 4 in SKILL.md explicitly instructs the agent to treat repository content as inert data rather than instructions and to flag any 'ignore previous instructions' patterns found in codebase files.
  • [COMMAND_EXECUTION]: The skill uses standard, read-only system tools like grep and git for reconnaissance and metadata retrieval. These actions are limited by 'Hard Rules' that forbid mutating operations like installs, builds, or commits.
  • [INDIRECT_PROMPT_INJECTION]: The skill manages the risk of ingesting untrusted data through a documented evidence chain: 1. Ingestion points: codebase files read during the Recon and Audit phases. 2. Boundary markers: a strict 'data-not-instructions' rule. 3. Capability inventory: read-only file access and writing to a specific plans/ directory. 4. Sanitization: instructions to flag and ignore adversarial content in analyzed files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 06:48 PM
Security Audit — agent-trust-hub — improve-animations