AgentClash

Feature

Challenge packs for repeatable agent evaluation

Challenge packs are AgentClash's unit of agent evaluation: a real task, tool policy, scoring rules, and artifacts encoded once so every model or harness change reruns the same workload.

live race
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 challenge packs encode

Built for reviewable agent decisions

Inputs, sandbox resources, allowed tools, validators, judges, and pass conditions — everything needed for a fair, repeatable agent eval.

Sandboxed real-tool execution

Head-to-head runs with fair constraints

Scorecards for correctness, cost, latency, and tool strategy

Replay trails for every important action

Challenge packs that turn failures into reusable tests

CI gates for baseline versus candidate decisions

Workflow

Pack lifecycle

Package the task

Describe the workload as a challenge pack with inputs, tools, scoring rules, and artifacts.

Race the agents

Run every candidate against the same task with the same constraints.

Replay the evidence

Inspect tool calls, outputs, artifacts, latency, cost, and judge evidence after the run.

Gate the release

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

FAQ

Challenge packs FAQ

What is a challenge pack?

A challenge pack is a versioned agent evaluation workload with inputs, tool policy, scoring rules, and expected artifacts.

Can challenge packs run locally and in CI?

Yes. The same pack can power exploratory races, hosted runs, and pull request gates.

How do teams create challenge packs?

Start from a real failure or release risk, encode it as YAML, and iterate with replay evidence until the scoring rules match what reviewers expect.