AgentClash

Feature

Synthetic dataset generation inside AgentClash

Generate eval-ready examples from pinned seeds without leaving your workspace. Choose fast prompt-only expansion or Agentic Self-Instruct with weak-vs-strong judge filtering.

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

Generation strategies

Built for reviewable agent decisions

Fast Self-Instruct adds volume quickly. Agentic Self-Instruct runs weak and strong solver rollouts with acceptance policies tuned to the useful difficulty zone.

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

From generation to gates

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

Synthetic generation FAQ

Where do I start generation in AgentClash?

Open Workspaces, Datasets, your dataset, then Synthetic generation in the UI or use agentclash dataset generate from the CLI.

What happens to rejected examples?

Rejected rows are stored with reason codes and solver attempts so you can review why the judge declined them.

Can I export for fine-tuning from AgentClash?

AgentClash optimizes for eval formats. For SFT, DPO, and Hugging Face export, use the DataSmith Python SDK on the same seeds.