AgentClash

Glossary

What is a challenge pack?

Challenge packs are AgentClash's unit of repeatable agent evaluation. Encode the task once so every model, prompt, or harness change reruns the same workload with the same constraints.

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

What packs contain

Built for reviewable agent decisions

Inputs, tool policy, sandbox resources, validators, judges, artifacts, and pass conditions that together define a fair eval or regression test.

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 pack to gate

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

Challenge pack FAQ

How is a challenge pack versioned?

Packs are versioned YAML bundles in your workspace. Pin a version for benchmarks and CI so comparisons stay reproducible.

Can one pack power both benchmarks and CI?

Yes. The same frozen pack can back a public benchmark eval and an internal release gate once your team trusts the scoring rules.

Where are examples?

See example packs in the repository and the challenge pack reference docs for field-by-field authoring.