Skip to content

Releases

v0.4.0

Choose a tag to compare

@goshtasb goshtasb released this 11 Nov 00:03
94a7140

A first review on any setup: it checks which AIs are ready, runs with those, and finishes even when a seat fails.

Added

  • council review checks every AI CLI (Claude, Gemini through agy, Grok) before it spends anything: a one-word prompt in an empty folder with no tools. Each line says ready, not installed, not signed in or out of quota, with the command that fixes it. --no-check skips it.
  • council doctor works outside a set-up repository: it runs the same check and says which seats a review would use.
  • A seat that fails partway (usage limit, lost login, turn limit, error) moves to another model or AI for the rest of the review. A lost login drops that AI; any other failure drops only that model, so Claude's Engineer seat falls back from opus to sonnet before leaving Claude. A seat only moves to an AI already in the review's line-up: with a .council/config.yml, one its seats name or one listed under council.failover_providers. The review lists each swap under "Gaps in this review" and records the line-up in seats.json. When no AI is left for the Chief Architect, the review finishes without its rulings.
  • Reviews say who actually held each seat. A review on Claude alone is labeled single-vendor and names its models.
  • When the Chief Architect overrules every objection, the review says so instead of reporting a stalled review with nothing found.
  • A review ends with a verdict: PASS when nothing that blocks is left, CHANGES REQUESTED when an objection stands, needs a measurement, or was never ruled on. It replaces STALLED, which every review showed because a review almost never reaches the PRD's zero-concern sign-off.
  • The Chief Architect can rule an objection MINOR: right, but it should not block (naming, style, wording, a nice-to-have). Minor notes are listed after the blocking objections and never hold up a review. STANDS now means a defect that should block.

Changed

  • council review runs one quick round by default: the Lead PM's brief, the Chief Engineer's objections, and the Chief Architect's ruling on each. --debate runs the earlier two-round debate. --rounds now takes 1 to 4.
  • A seat in .council/config.yml whose AI is not ready moves to a ready one for that review only; the file is not changed.
  • Every review, the debate included, ends with the Chief Architect ruling on whatever is still open, so the verdict rests on its rulings.
  • The Chief Engineer is told to object to defects in the change and the code it touches, not to the wording, numbering or test plan of the Lead PM's brief. Those objections were why a second round never converged.
  • The GitHub Action's outcome output is now PASS, CHANGES_REQUESTED, NO_VERDICT or SKIPPED, instead of AWAITING_APPROVAL or STALLED. A workflow that checks for the old values needs updating.
  • A login failure is recognised only from what a seat's CLI prints about its own login: Claude Code's own login lines on stdout only when they are the whole of it, and other login words, a bare 401 or 403 only on stderr. A model's answer that talks about 401s or logins no longer drops its AI. The Action's login error no longer assumes the seat was Claude.
  • The GitHub Action still passes --rounds 2, so it keeps the two-round debate. It skips the setup check (--no-check): a rejected token still ends the run with the Action's own "rejected the login" error, now also when the seat's failure is reported as "is not signed in". But a seat that fails mid-run now falls back to another Claude model and the comment lists the gap.

v0.3.2

Choose a tag to compare

@goshtasb goshtasb released this 08 Nov 19:24

The Gemini seat works again on current Antigravity builds, and every seat treats notes pointing outside the repo as unverified claims.

v0.3.1

Choose a tag to compare

@goshtasb goshtasb released this 07 Nov 20:50
5b54f48

Run the council on every pull request, on your own Claude plan or your own API key.

Added

  • A GitHub Action (uses: goshtasb/OpenCouncil@main) that reviews each pull request and posts the strongest file:line objections as one comment, with a link to the full replay. It takes your plan token from claude setup-token (claude_code_oauth_token) or your own Anthropic API key (anthropic_api_key, recommended for team repos), edits its earlier comment instead of adding more, and skips fork, draft and login-less runs with a notice. See "Review every pull request" in the README.
  • council review prints which login each seat uses (your plan or an API key) before the seats start, and warns when ANTHROPIC_API_KEY would bill a key instead of your plan. council doctor names it for each seat too.
  • council review --comment <file> writes a PR-comment version of the review, and --replay-url <url> links it to the full replay.
  • With a Gemini API key (or Vertex turned on) and gemini-cli installed, the Lead PM seat runs on gemini-cli, since Antigravity cannot use a key.

Fixed

  • A seat CLI that fails is reported by name with its last lines of output, not its whole command line.
  • file:line locations keep a leading dot (.github/...).

v0.3.0

v0.3.0 Pre-release
Pre-release

Choose a tag to compare

@goshtasb goshtasb released this 07 Nov 16:32
v0.3.0

v0.2.1 — first outside contributors

Choose a tag to compare

@cursor cursor released this 06 Nov 19:24

A small release, and the first with outside contributors. Thanks to @O1sumitkumar and @Doomguy-BFG9000.

Added

  • council backlog list --json prints the backlog as a JSON array (id, title, status, priority, session; session is null until a session takes the item), and [] for an empty backlog, so scripts can read it (#14).
  • CODE_OF_CONDUCT.md (Contributor Covenant 2.1), a "Your first contribution" path and an AI-assisted contributions policy in CONTRIBUTING.md, and issue routing that sends questions and run reports to Discussions (#13).

Changed

  • On native Windows every council command now stops at once with a message pointing to WSL, instead of failing later inside a gate or project command (#15).
  • README states what a run costs right under the install line.

v0.2.0 — one CLI is enough

Choose a tag to compare

@goshtasb goshtasb released this 15 Oct 19:28
436201b

The entry ticket drops from three CLIs to one, and the project ships through npm, the Claude Code plugin marketplace and the skills ecosystem.

Added

  • Published as open-council on npm: npx open-council init. Install-time scripts removed from the package.
  • Claude Code plugin: /plugin marketplace add goshtasb/OpenCouncil, /plugin install open-council@open-council, then /open-council:council "<task>" drives backlog → open → deliberate, stops for the human approval, then handoff → run.
  • Agent skill: npx skills add goshtasb/OpenCouncil installs the same council skill for other agents.
  • Solo mode. council init detects which provider CLIs are installed and assigns seats from what it finds.
    With only Claude Code present every seat runs on Claude Code (Lead PM on sonnet, Chief Engineer on opus,
    Chief Architect on sonnet with all tools disabled). --solo and --full force a plan; --force rewrites
    an existing config.yml and keeps customized personas, contracts and standards.
  • Quick profile. council init --quick or council open --quick: at most 4 rounds, tie-break from round 3,
    one sign-off revision, the six starred review points, and documents capped at 1,200 words.
  • Partial result on a stall. A council that stalls, or dies on a seat failure, writes
    stalled-draft-and-objections.md (task, last draft, open objections, unresolved rulings and concerns) into the
    session instead of leaving nothing. It is never approvable.
  • Resubmission rule for the Architect review. On a resubmitted document the Chief Architect sees its previous
    review and the council's response, and a concern that could have been raised on the previous version and was not
    is waived. Found on the first live solo run: the Engineer ratified every revision while the Architect raised a
    different advisory concern each round until the budget ran out.
  • Early arbitration. When open objections have not dropped for no_progress_rounds (default 3) consecutive
    rounds the Chief Architect arbitrates immediately instead of at tiebreak_round.
  • One retry on transient seat failures. An empty reply, a dropped connection or a transient 5xx from a seat
    is retried once after 5 s; timeouts are not retried.
  • The execution prompt names the files the dependency install step created and tells the Chief Engineer to leave
    them uncommitted (observed: an unrequested "commit npm lockfile" commit on the first solo-mode PR).
  • council doctor prints the one command that fixes each failing seat (install, login or model list) and
    suggests council init --solo --force when only Claude Code works.
  • ROADMAP.md, this changelog, and GitHub Discussions for ideas, questions and council run reports.
  • Issue and pull request templates, and a security policy.
  • docs/office.gif: the pixel office replaying the seat states of the session that built PR #1; docs/showcase/ case studies.

Changed

  • Project renamed from Open Councilmen to Open Council.
  • README rewritten outcome-first: one-liner, one-line install, five-minute run, how it differs from an LLM council, showcases, honest limits.
  • The Charm announcement moved out of the README into ROADMAP.md.
  • CLI --version reads package.json.

Fixed

  • This repository's own .council/ config is no longer tracked in git.
  • The stale open-councilmen-0.1.0.tgz is no longer tracked; packed tarballs are ignored everywhere.
  • The repository description and homepage on GitHub pointed at the pre-rename name.
  • CI runs one macOS job instead of three, after repeated 15-minute runner-acquisition failures.

Install

npm install -g open-council-cli      # commands: council, open-council
npm install -g https://github.com/goshtasb/OpenCouncil/releases/download/v0.2.0/open-council-cli-0.2.0.tgz

Claude Code plugin: /plugin marketplace add goshtasb/OpenCouncil then /plugin install open-council@open-council. Skill: npx skills add goshtasb/OpenCouncil.

Showcase of the first solo-mode run: council-playground PR #1, written up in docs/showcase/council-playground.md.

Charm (latest build)

Charm (latest build) Pre-release
Pre-release

Choose a tag to compare

@github-actions github-actions released this 04 Oct 03:40

Built from 2dd88a7 on feat/charm-bridge. Open this page on your phone and tap council-charm.apk to install.

v0.1.0 — a council that has to agree before you approve

Choose a tag to compare

@goshtasb goshtasb released this 18 Sep 14:47

Open Council runs three AI CLIs as a council with separated powers, and hands you exactly one decision.

  • Lead PM (Gemini, via the Antigravity CLI) researches the repository and writes a Product Brief + PRD.
  • Chief Engineer (Claude Code) verifies every claim against the actual code, objects with file:line evidence, and refuses to ratify what it cannot test.
  • Chief Architect (Grok) arbitrates deadlocks, rules on anything the standards leave undefined, and audits the spec against a 12-point industry-standards checklist.
  • You approve the finished PRD. That is the only human step, and nothing is written before it.

A document only reaches you when all three seats have signed off the byte-identical file: the Engineer with zero objections, the Architect with zero concerns, the Lead last. finalize and approve both re-check that against the SHA-256, and the PRD is handed over as a PDF. After approval the Chief Engineer implements in an isolated clone, the harness merges your base branch, runs your verification gates, and only then pushes and opens the pull request. Anything that fails is reported as BLOCKED — never as done.

Verification gates

Standards are checked by commands, not by a model's word:

verification:
  gates:
    - name: "test"
      command: "npm test"
      standard: "12-point #4 Testing & Release Engineering"
    - name: "dependency-audit"
      command: "npm audit --audit-level=high"
      standard: "12-point #3 Software Supply Chain"

Every run writes an evidence record to the session and a gate summary into the pull request body.

It has done this to itself

The council specified a feature for this repository, Claude implemented it, the gates passed, and it opened PR #1. That PR then failed CI on a semantic merge conflict the harness had not checked for — so the harness now merges the base branch before its gates run, and the feature it wrote (council gates) is in this release.

Honest limits

  • A full council takes roughly 10–40 minutes and uses real subscription quota.
  • Live runs so far are single digits; one in five stalled without producing a document.
  • macOS and Linux only (gates run through a POSIX shell). CI covers Node 18/20/22 on Ubuntu and macOS.
  • gemini-cli is unusable on current individual Gemini tiers — use agy for Gemini seats.
  • The Ollama adapter has never been run.
  • Neither Antigravity nor gemini-cli can be stripped of tools, so neither may hold the zero-tool Architect seat; doctor warns about this.

Getting started

npm install -g https://github.com/goshtasb/OpenCouncil/releases/download/v0.1.0/open-council-0.1.0.tgz
council init
council doctor
council backlog add "..." --priority 1
council open my-change --item 001 && council deliberate <session>

Contributions welcome — see CONTRIBUTING.md. Provider adapters, gate recipes for other stacks, and reports of councils that went wrong are the most useful right now.