4-step workflowverification adversarial sub-agents qa verdict guided-flow Agent QA & Security

Stand Up an Adversarial Verification Agent

Stop trusting your agent's own 'it works': dispatch a read-only verifier that must run real commands, probe adversarially, and return an evidence-backed PASS/FAIL verdict.

Run this as a guided flow on Flows — verify gates and recovery prompts included →

Flows@flows12d ago

0 runs0 forks0

Chain

Inputs

Step 1

flow-adversarial-verification-agent-verdict-1-define-the-verifier-with-hard-read

Step 2

flow-adversarial-verification-agent-verdict-2-feed-it-the-real-task-and-require-

Step 3

flow-adversarial-verification-agent-verdict-3-add-adversarial-probes

Step 4

flow-adversarial-verification-agent-verdict-4-gate-completion-on-the-verdict

Final output

Workflow inputs

No inputs required — hit Run to stream each step in order.

Each step streams from its model and feeds the next. BYOK — you pay your provider directly. Add a key in Settings if a step needs one.

Step 1 · Constraints in the tool pool, not just the prompt.

flow-adversarial-verification-agent-verdict-1-define-the-verifier-with-hard-read

Claude Sonnet 4.6→ {{step_1}}

Awaiting upstream steps…

Step 2 · Verification quality is bounded by input honesty and output proof.

flow-adversarial-verification-agent-verdict-2-feed-it-the-real-task-and-require-

Claude Sonnet 4.6→ {{step_2}}

Awaiting upstream steps…

Step 3 · The happy path passing is where real verification begins.

flow-adversarial-verification-agent-verdict-3-add-adversarial-probes

Claude Sonnet 4.6→ {{step_3}}

Awaiting upstream steps…

Step 4 · A verdict nobody acts on is decoration.

flow-adversarial-verification-agent-verdict-4-gate-completion-on-the-verdict

Claude Sonnet 4.6→ {{step_4}}

Awaiting upstream steps…

About this workflow

Stop trusting your agent's own 'it works': dispatch a read-only verifier that must run real commands, probe adversarially, and return an evidence-backed PASS/FAIL verdict.

What happens when you run it

1
Constraints in the tool pool, not just the prompt.
Runs flow-adversarial-verification-agent-verdict-1-define-the-verifier-with-hard-read on Claude Sonnet 4.6
2
Verification quality is bounded by input honesty and output proof.
Runs flow-adversarial-verification-agent-verdict-2-feed-it-the-real-task-and-require- on Claude Sonnet 4.6
3
The happy path passing is where real verification begins.
Runs flow-adversarial-verification-agent-verdict-3-add-adversarial-probes on Claude Sonnet 4.6
4
A verdict nobody acts on is decoration.
Runs flow-adversarial-verification-agent-verdict-4-gate-completion-on-the-verdict on Claude Sonnet 4.6

Best for

•Recurring tasks you currently copy-paste between tools.
•When a single prompt isn't enough — you need a chain where output(n) feeds prompt(n+1).
•Fork and adapt if your inputs differ materially from the author's use case.

Models used

Claude Sonnet 4.6