Markdown source

Evidence-Gated Agent Workflow

Pattern

A pattern for letting agents act only when the workflow can name the source evidence, acceptance gate, and review path attached to the action.

When To Use

Use this when an agent can change code, call tools, retrieve external context, or produce a decision that another system may rely on.

Implementation Moves

• Define the task boundary and the evidence required before the agent starts.
• Attach eval or policy checks that can stop, route, or downgrade the result.
• Keep source links, traces, diffs, screenshots, or transcript/slide references with the output.
• Require human approval when the evidence bundle is incomplete or the action crosses a high-risk boundary.

Source Evidence

• agent work needs runtime boundaries - Claim synthesis
• evals are operational gates - Claim synthesis
• agent eval gate - Harness synthesis
• coding agent code review loop - Harness synthesis
• agent security - Topic synthesis

Evidence Boundary

This is a reusable pattern synthesized from linked conference evidence. Treat it as an engineering abstraction, not as an official event claim or a direct quote.