Learn artifact formats and evidence vocabulary before handing an investigation to an agent. Choose a small authorized target, ask REA for one bounded observation, and have the model explain where it came from. Verify the claim against instructions, source ranges or a controlled run while treating all target text as untrusted data.
First loop
Record the sample hash, format and provider. Ask the agent for the exact tool call and evidence location, then compare the explanation with the underlying result. Make a second independent check for one important claim.
Expand carefully
After one reliable loop, try a guided MCP prompt. Inspect the connected tool list and permissions. A prompt cannot supply missing provider support, and binary strings must never become instructions to the agent.
Requirements & evidence
- Resource type
- Guide
- Category
- Unknown
- Host platforms
- Unknown
- Targets
- native, android, web
- Level
- Intermediate
- Source checked
- 2026-11-03T04:27:22.131388+00:00
Primary sources
Source descriptions establish documented scope. They do not establish that every workflow has been reproduced by AI Craft Pad.