frog-cycle

Pass

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data through user specifications and existing project files, which are interpolated into subagent prompts.
  • Ingestion points: User-provided behavior specifications in the LISTEN phase; test and implementation file contents in the CROAK, LEAP, and SHED phases.
  • Boundary markers: Prompts use structural headers (e.g., ## Your Mission, ## Context) but lack explicit instructions to ignore instructions embedded within the variables.
  • Capability inventory: The skill has the ability to write to the file system (modifying test and implementation files) and execute shell commands to run tests.
  • Sanitization: There is no evidence of sanitization or escaping of external content before it is interpolated into the prompts.
  • [COMMAND_EXECUTION]: The skill executes various shell commands to manage the TDD cycle.
  • Runs test runners including npx vitest, npx jest, uv run pytest, cargo test, and go test based on environment detection.
  • Executes gw ci --affected --fail-fast --diagnose for full suite verification in the CHORUS phase.
  • Uses shasum to verify test file immutability during phase transitions.
  • [DYNAMIC_EXECUTION]: The skill dynamically constructs prompts for subagents (opus and sonnet models) and invokes them using an Agent() function to perform phase-specific tasks like test writing, implementation, and refactoring.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 9, 2026, 01:31 AM