Capabilities
Four patterns, each documented on a static page and backed by a workflow you can re-run locally.
Long-context code review
K3 reads the whole target and returns evidence-backed findings; MiniMax-M3 cross-checks for whatever it missed.
AgentSwarm exploration
Parallel read-only subagents fan out across disjoint scopes, then a lead agent synthesizes one verified answer.
Quota-aware fallback
Entitlement and 402 errors classify as quota, so chains degrade from Kimi to MiniMax-M3 without human intervention.
Prompt-to-site
Static webpage concepts produced as safe, reviewable artifacts entirely inside the lab boundary.
The agentic loop
Gates are structural: downstream agents never spawn until the upstream gate clears. The same loop reviewed this page before it reached you.
- Dispatch.
A single task contract fans work out to heterogeneous workers — kimi, MiniMax, or claude profiles.
- Review.
K3 examines the diff or files and returns schema-validated findings, never free-form prose.
- Cross-check.
A second vendor re-verifies each finding and hunts for omissions; corroborated issues rank first.
- Gate.
Results halt at
gate: pass. A human approves before anything further happens.
Run the workflow yourself
The pages are static; the orchestration is real. On the sample fixture, expect a halt at gate: pass — the fixture ships a deliberate high-severity bug.
node bin/aflow.js workflow run top-review-lab --variant strict
# or the two-model corroboration loop:
aflow flow run crosscheck --diff HEAD~1 "review this change"Safety boundaries
- No credentials, transcripts, or session state in any demo page.
- No mutation outside
experiments/skills-workflows-hooks-lab/. - Static HTML/CSS only — no backend, no trackers, no animation budget.