luna-local-review-loop

Pass

Audited by Gen Agent Trust Hub on Aug 23, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local shell scripts to manage task lifecycles, registry states, and process coordination. These scripts utilize standard Unix utilities including ps, kill, stat, and jq to maintain a local execution environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill acts as an orchestration layer for processing external task prompts and repository data.
  • Ingestion points: Task instructions are read from caller-provided files (e.g., scripts/run-worker.sh reads PROMPT_FILE).
  • Boundary markers: The system automatically wraps task prompts with a mandatory "sole worker" instruction to prevent sub-agent spawning or delegation.
  • Capability inventory: The skill can manage local processes, read Git metadata, and modify its internal JSON-based registry.
  • Sanitization: Structured worker results are validated against a formal JSON schema to ensure data integrity before being accepted by the parent agent.
  • [SAFE]: The skill follows security best practices, such as using a restrictive umask (077), explicitly setting private file permissions (chmod 0600), and verifying repository identity through Git administration evidence and checkout seals.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 23, 2026, 06:40 PM
Security Audit — agent-trust-hub — luna-local-review-loop