AgentClash

Glossary

What is Agentic Self-Instruct?

Agentic Self-Instruct generates synthetic training examples by proposing tasks, running weak and strong solvers, and accepting rows only when the strong path succeeds and the weak path struggles.

live eval
gate: pass

Candidate

92correct patch, low cost

Baseline

88stable reference run

Control

73missed edge case

replay timeline

1loaded task inputs and tool policy
2ran sandbox actions and captured artifacts
3scored trajectory and validator evidence
4attached scorecard and release verdict

ci verdict

Candidate clears release gate

Correctness improved, latency within budget, and required artifacts were preserved for review.

agentclash run create --follow

How it differs from self-instruct

Built for reviewable agent decisions

Classic self-instruct prompts a model for more examples. Agentic Self-Instruct adds solver rollouts and a judge so difficulty is measured, not assumed.

OpenTelemetry-compatible trace import

Pinned datasets and golden test cases

Baseline versus candidate regression checks

Replay trails for tool calls, outputs, and artifacts

Scorecards for correctness, cost, latency, and evidence

CI gates for prompt, model, RAG, and tool changes

Workflow

Typical roles

Import the evidence

Start from OpenTelemetry traces, curated datasets, support transcripts, or a real failure your team already saw.

Pin the baseline

Record the current accepted behavior so every prompt, model, RAG, or tool change has a fair comparison point.

Replay the evidence

Inspect tool calls, outputs, artifacts, latency, cost, and judge evidence when a candidate gets worse.

Gate the release

Compare candidate and baseline runs, then fail CI before a regression reaches users.

FAQ

Agentic Self-Instruct FAQ

Who popularized Agentic Self-Instruct?

Meta FAIR's Autodata paper formalized weak-vs-strong agentic self-instruct for synthetic data generation and meta-optimization.

What is the useful difficulty zone?

Examples where a strong solver passes and a weak solver fails, indicating the row can teach the weak model something new.

Where can I run it?

DataSmith (pip install datasmith) for local export, or AgentClash workspaces for hosted generation tied to eval gates.