checkpoint-iterate

Pass

Audited by Gen Agent Trust Hub on Aug 2, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill performs local shell operations including Git commands for metadata retrieval (git rev-parse) and the bats test runner for code validation. These actions are standard for development utilities and restricted to the project environment.
  • [PROMPT_INJECTION]: The iteration loop processes user instructions to drive code modifications, representing a surface for indirect prompt injection.
  • Ingestion points: User-provided descriptions of next steps, the $ARGUMENTS parameter, and the session ledger file (docs/cr/{CR_ID}-iterate.md) which is read back during session resumption.
  • Boundary markers: The templates/ITERATE.md template uses specific markdown headers to distinguish recorded attempts and evidence.
  • Capability inventory: The skill is authorized to write to the docs/cr/ directory, execute local commands, and invoke the /checkpoint-commit skill.
  • Sanitization: The skill follows user-driven instructions without additional input filtering, which is consistent with its role as an interactive developer assistant.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 2, 2026, 03:59 PM
Security Audit — agent-trust-hub — checkpoint-iterate