stay-within-limits

Pass

Audited by Gen Agent Trust Hub on Sep 11, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes shell scripts to manage Claude API usage snapshots.
  • capture-rate-limits.sh and select-review-profile.sh use standard Unix utilities (jq, mktemp, chmod) to process and store data.
  • Updates to the usage snapshot are performed atomically and file permissions are restricted to the current user (mode 0600).
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from the Claude API environment.
  • Ingestion points: Reads Claude status information from standard input in capture-rate-limits.sh.
  • Boundary markers: The input is expected to be JSON; no specific boundary markers are used for the raw stream.
  • Capability inventory: Operations are restricted to local file system reads and writes (e.g., ~/.claude/rate-limits.json). No network capabilities or unsafe evaluation functions are present.
  • Sanitization: Uses jq to strictly validate that the input contains specific numeric fields before storage, preventing the processing of arbitrary or malicious payloads.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 11, 2026, 02:00 AM
Security Audit — agent-trust-hub — stay-within-limits