keep-going

Pass

Audited by Gen Agent Trust Hub on Aug 19, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill contains instructions that explicitly command the agent to override its own prior reasoning and decision-making. Specifically, it tells the agent to "discard" its prior conclusions and that "the user's instructions outrank your own last message."
  • [INDIRECT_PROMPT_INJECTION]: The skill creates a behavioral surface where the agent is instructed to interpret user phrases like "keep going" or "continue" as a mandate to bypass its usual cautious stopping points or requests for clarification, potentially leading to actions taken with less oversight.
  • Ingestion points: Processes user phrases "keep going", "continue", "unstuck yourself", "why did you stop", and the /keep_going command.
  • Boundary markers: None identified for external data; the instructions apply to the agent's internal state.
  • Capability inventory: Directs the agent to "chain actions, one after another" and "execute it right now" using available tools.
  • Sanitization: The skill relies on the agent's ability to "prove" a blocker exists to trigger a stop, which is a subjective internal evaluation.
  • [SAFE]: The skill defines clear safety boundaries by requiring the agent to stop for "irreversible or externally-visible actions" such as spending money, sending external messages, or performing production writes.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 19, 2026, 07:20 PM
Security Audit — agent-trust-hub — keep-going