Skip to content

v0.15.0

Choose a tag to compare

View all tags
@arminzou-agent arminzou-agent released this 27 Oct 03:54
v0.15.0
4cb63f8
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Packaged subagents become optional per project, global settings move into one file, and workers record their own results reliably on every runtime.

⚠ Breaking changes

Before After
dotbrain agents list/link dotbrain subagents list/link (custom subagents only)
Packaged subagents always present (Claude Code, everywhere) Listed per project in project.yaml packaged-subagents; delivered only to wired checkouts
dotbrain:<role> in Claude Code dotbrain-<role> on Claude Code and Codex
skills/skills.yaml, agents/agents.yaml, targets One config.yaml (version 4) with profiles, skills, subagents

Upgrade: run dotbrain bootstrap once to migrate global settings (originals are backed up), then dotbrain refresh in each wired checkout to record and deliver its packaged subagents. Legacy targets for runtimes other than Claude Code and Codex are dropped with a report.

Highlights

  • Optional packaged subagents. Each project lists the roles it wants (worker, researcher, reviewer, verifier); edit the list and refresh to opt out. Workflow skills now carry each role's contract, so they run with or without packaged subagents.
  • Agent working folder. wire and refresh keep .dotbrain-tmp/ git-ignored in every checkout. Workers and reviewers write their comment files there, so results reach Beads intact on every runtime.
  • Workflow skills. Skills start from plain requests, curate retrospectives from work records and apply findings before work, and record acceptance evidence kinds and exemptions.
  • Core skill evals run on Claude Code and Codex with Promptfoo.

Fixes

  • refresh warns when global settings still need dotbrain bootstrap.
  • bootstrap migrates legacy global profile configuration.
  • Codex project skills move to .agents/skills.
  • Windows: runtime CLIs installed as .cmd are found.
  • Worker dispatch routes through run-execution on both runtimes.
  • Test suites bound heavy checks and pytest worker concurrency; sync tests are stable.

For maintainers

The runtime delivery UAT is now a no-model, no-key container suite that checks install, discovery and hooks for all six runtimes with real binaries. See .github/RELEASING.md.

Changelog

⚠ BREAKING CHANGES

  • Packaged Claude subagents now require a wired checkout. Run dotbrain refresh to deliver the roles selected in project.yaml under dotbrain- names. Global linking manages custom subagents only.

Features

  • evals: run core skill evals on Claude Code and Codex with Promptfoo (#60) (9b51ad2)
  • make packaged subagents optional per project (#69) (da610b1)
  • skills: curate retrospectives from work records and use findings before work (#65) (7965b7a)
  • skills: let workflow skills start from plain requests (#63) (4291008)
  • skills: record acceptance evidence kinds and exemptions (62ac0bd)
  • wiring: give agents an ignored .dotbrain-tmp folder for comment files (#73) (e5114af)

Bug Fixes

  • bootstrap: migrate legacy global profile configuration (#66) (95c330c)
  • codex: migrate project skills to .agents/skills (#62) (e3e2fc0)
  • probe .cmd runtime CLIs on Windows and add an opt-in delivery UAT suite (#59) (949f71c)
  • refresh: warn when global settings still need migration (e98d044)
  • skills: keep static-only evidence exclusive (295903d)
  • skills: route worker dispatch through run-execution on both runtimes (#64) (62c8567)
  • stabilize sync tests and recover retro candidates (#67) (6f152da)
  • tests: bound heavy checks and pytest worker concurrency (#70) (9a9ef4c)
Assets 2
Loading