| name | safety-check |
| description | Manually check whether a specific browser action is safe to execute autonomously or requires human approval. Use when unsure about an action's risk level. Invoked with /safety-check followed by the action description.
|
Safety Check Skill
You have been invoked to assess a single browser action for safety.
What to do
- Extract the action description from the invocation arguments
- Delegate immediately to the
action-validator subagent
- Return the full risk assessment to the user in a readable format
Delegation prompt to action-validator
Assess the following browser action for risk:
ACTION: [extracted action description]
Return: RISK level, REASON, FLAGS, and RECOMMENDATION.
Present result to user
Format the result clearly:
Safety Assessment
─────────────────
Action: [action]
Risk: [HIGH / MEDIUM / LOW]
Reason: [one-line explanation]
Flags: [list or "none"]
Verdict: [proceed / proceed-with-caution / human-required / block]
If HUMAN-REQUIRED or BLOCK:
- Explain what the user should do manually instead
- Offer to continue the task after the user completes the step
