Guildhall 0.4.0

Summary

Guildhall 0.4.0 is the release where the automation story becomes real for a bounded lane of project work.

The system now has live proof that narrow, low-blast-radius cleanup tasks can complete end to end across:

• worker implementation
• reviewer handoff
• gate-check adjudication
• terminal merge truth

This is not a claim that Guildhall can autonomously finish every task shape. It is a narrower and stronger claim: a real class of project work now survives the full queue without manual babysitting.

What changed

• Worker reclaim and handoff state are much more stable.
• Reviewer output is more pragmatic and less absolutist.
• Coordinator/gate adjudication can now demote unrelated repo-red and broad doctrinal spillover instead of blocking narrow tasks incorrectly.
• Terminal task outcomes now expose truthful merge/publish summaries.
• The product UI is clearer about runtime state, review signal, and coordinator responsibilities.

Live proof behind this release

Looma/Knit narrow-cleanup cohort:

• task-015
• task-016
• task-017

All three completed with terminal merge truth, and task-017 was replayed live on the fresh build through:

ready -> in_progress -> review -> gate_check -> done

Supported claim

For 0.4.0, the honest supported claim is:

Guildhall can autonomously complete a proven class of narrow engineering tasks end to end.

Good candidates right now:

• one-file cleanups
• small test/code hygiene fixes
• low-blast-radius type or lint follow-ups
• small local repairs with clear task scope

Not yet proven by this release

Still outside the proven lane for this release:

• broader feature work
• cross-cutting refactors
• multi-surface behavioral changes
• medium-complexity tasks with large review/gate blast radius