super-ralph-wiggum

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a feedback loop that ingests data from external files and previous assistant outputs to drive subsequent iterations.
  • Ingestion points: The skill reads task definitions from prd.json, session history from progress.txt, and instructional content from various template files (e.g., templates/feature-prd.md). These are injected into the agent's prompt during setup and at the start of each loop iteration (see scripts/setup-ralph-loop.sh and scripts/stop-hook.sh).
  • Boundary markers: The injected content uses standard Markdown headers for structure, but lacks explicit security delimiters or "ignore embedded instructions" warnings for the external data sections.
  • Capability inventory: The agent platform provides capabilities for file system modification (read/write) and shell command execution to implement requested features and run tests.
  • Sanitization: There is no sanitization or filtering of the content read from prd.json or progress.txt before it is interpolated into the agent's prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 06:34 PM