Start with the project README for a first run. Use this index when extending a runtime or reproducing a result.
| Task | Guide |
|---|---|
| Install; use the CLI, SDK, or authored workflows | Package guide |
| Set providers, models, teams, and context policy | Configuration |
| Choose the default or a concrete agent profile | Base and Single2 |
| Compare and adopt two isolated candidates | Duo · Chinese translation |
| Run ordered work with checks, repair, and continuation | Weave |
| Reconstruct a hidden program through active exploration | Reconstruct |
| Control workflow calls; integrate evaluation usage, cancellation and cleanup | Evaluation runtime |
| Inspect model capabilities and effective profiles offline | Model inspection |
| Record native test observations and parse Git patches | Test evidence |
| Add on-demand instructions | Skills |
| Understand context, budget allocation, and steering | Runtime design |
| Task | Guide |
|---|---|
| Prepare a contribution and follow collaboration rules | Contributing |
| Find tests; run Python, DOM, and browser checks | Contributor checks and test map |
| Use launchers, diagnostics, and experiment-card generation | Scripts |
| Measure tool-loop processing and progress recovery | Tool-loop measurements |
| Publish a tagged GitHub release | Releasing |
| Report a vulnerability privately | Security |
Benchmark results reports the recorded cross-harness comparison and paired Duo–Base test. Adherence separates declared team conditions from measured collaboration. The collaborating-team study and experiment cards live with their generators and checks in OpenCollab-Eval. Benchmark datasets, runs, and remote execution belong to OpenCollab-Eval.
The release provenance and source coverage preserve the October 1 integration ancestry and its original path dispositions. The contributor architecture guidance links the archived June survey and explains how to interpret its findings. Historical snapshots retain their original revisions; current rules live in the contributor and runtime guides above.
Migration guides describe the named release, not current signatures or defaults.
| From / to | Guide |
|---|---|
| 0.7.x → 0.8.0 | Base profile and Single2 defaults |
| 0.6.x → 0.7.0 | Native Bash testing, evaluation runtime, and subsequent workflow changes |
| 0.5.x → 0.6.0 | Public API, teams, workflows, and lifecycle |
| Earlier versions → 0.5.0 | Runtime failures, isolation, configuration, and budgeting |