Skip to content

release: Team Avatar v1.0.0, a universal agent orchestrator - #57

Merged
tmcfarlane merged 14 commits into
mainfrom
release/v1.0.0
Oct 24, 2026
Merged

tmcfarlane merged 14 commits into
mainfrom
release/v1.0.0

Conversation

@tmcfarlane

@tmcfarlane tmcfarlane commented Oct 20, 2026 •

Copy link
Copy Markdown
Owner

Requested by Trent

Demo

team-avatar-demo-readme-720p.mp4

App screens are recreated from this repository's real runs and files. Narration is an AI clone of the author's voice.

The 2:10 launch film. The same video now opens the README.

Before: oh-my-cursor was a Cursor-only agent pack, and the Team Avatar work for Claude Code and Codex sat in three stacked PRs (#54, #55, #56).

After: one v1.0.0 release. The project is called Team Avatar, a universal agent orchestrator for Claude Code, Codex, and Cursor. The README says plainly that we took pstack, put Avatar characters on top, and that it probably made it worse but cooler. The benchmarks back that up: no plugin moves the SWE-bench score by a significant margin.

This PR carries every commit from #54, #55 and #56, plus a release commit that renames links and install commands to tmcfarlane/team-avatar, bumps the plugins to 1.0.0, adds the changelog entry, and adds screenshots.

See it work

These are real terminal sessions on a small demo repo with three failing tests. One prompt asks for one Katara per module in parallel. How they were made.

Claude Code. /team-avatar:kyoshi-formation starts three team-avatar:katara agents at once (agent panel at the bottom). 53 seconds, 3 of 3 passing.

Codex. $kyoshi-formation gives each Katara its own worktree and starts katara_cart, katara_slug and katara_paginate. The session log has three child rollouts pointing at the parent. 2 minutes 9 seconds, 3 passed, 0 failed.

What's in it

  • plugins/team-avatar: 58 skills owned by 9 characters (Aang, Sokka, Toph, Katara, Zuko, Iroh, Appa, Momo, Suki), one shared skills tree for Claude Code and Codex, and avatar-mode as the entry point.
  • A registry (team.json), a generator for roster tables, and a validator.
  • Evidence under docs/evidence/: 29 of 29 end-to-end cases on both tools, an A/B benchmark with hidden graders, and SWE-bench Verified Mini.
SWE-bench Verified Mini (50 tasks) Claude Code Codex
No plugin 48 42
Team Avatar installed 50 41
pstack installed 48 43

Every paired p-value is 0.5 or higher, and both plugins cost more per task.

How

Created release/v1.0.0 from the top of the stack (#56), so the history of all three PRs is kept as is. The old Cursor agents, rules and installers still ship unchanged. The repository rename to team-avatar happens right after this PR opens. GitHub redirects the old URLs.

🤖 Generated with Claude Code

tmcfarlane and others added 9 commits October 18, 2026 06:56
Team Avatar becomes a universal set of principal-level engineering
skills, with each discipline owned by a character from the show.
avatar-mode routes the work; the plugin installs from this repo as a
Claude Code marketplace and a Codex marketplace.

Adds 15 skills, 8 character agents, both marketplace manifests, a
zero-dependency validator wired into CI, and captured evidence of the
plugin loading. The Cursor install is unchanged.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Every pstack skill and playbook that Team Avatar was missing now has a
Team Avatar skill and a character owner. Suki joins the roster for
parallel squads and long autonomous runs.

team.json records each skill's owner and upstream source. A generator
rebuilds the roster tables from it, and the validator checks the
registry, the generated tables, references, links, manifests, and house
style. The plugin also gains the session-start hook, the avatar-setup
model sheet, the bun PR and orchestration tools with their tests, and a
NOTICE with the upstream MIT licenses.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The end-to-end tests showed Claude Code picking the right skill from the
route table and then working without loading it.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…d Codex

Adds a shared runner, fixture repos with planted defects, and the final
results: 29 of 29 on both runtimes, with every reply kept as evidence.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…rf fixes

The benchmark showed the session hook let Claude Code skip every skill on
small tasks, and Codex stop to ask instead of shipping a perf fix.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
… Codex

Seven tasks with hidden graders, three setups, three runs each. Results
from two rounds and an honest read of them are in
docs/evidence/team-avatar-bench.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…Avatar

Generates patches with headless Claude Code and Codex in fresh checkouts,
grades them with the official swebench harness, and audits every run for
answer leaks. Team Avatar did not change the resolved rate on either tool
(Claude Code 48 / 50 / 48 of 50, Codex 42 / 41 of 50) and costs more.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Rename the project to Team Avatar, point every link and install command at
tmcfarlane/team-avatar, bump the plugins to 1.0.0, credit pstack honestly in
the README and changelog, and add real Claude Code and Codex screenshots of
kyoshi-formation running three Katara subagents in parallel.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

tmcfarlane and others added 5 commits October 20, 2026 03:22
Add an "Other harnesses" section after Install. Every skill is a plain
Agent Skills SKILL.md, so OpenCode, Hermes Agent, OpenClaw, Muse Code and
other harnesses can load them. Document the skills CLI command, pointed at
plugins/team-avatar because the repository root also holds the Cursor
edition's skills, the folders each harness scans, and what a skills-only
install leaves out: the character subagents, the session hook, and the
helper tools.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Embed the 2:10 Team Avatar film (GitHub-hosted, so it plays inline) right
under the title block. The same video sits at the top of PR 57.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Root README:
- Move the origin story up front, ahead of the honest version, to match
  the launch film.
- Group every Cursor-edition section under "Team Avatar for Cursor" as
  subsections, merge its two install sections, and replace the
  Cursor-only contents list with a jump list.
- Fix stale facts: 9 slash commands (not 8), 21 bundled skills (not 19,
  and add the missing cursor-image-generation row), `cd team-avatar`
  after the rename, and the FAQ that sent Claude Code and Codex users to
  the old installer flags.
- Note Codex's multi_agent switch and avatar-setup in Install.
- Point the oh-my-opencode credit at its new name, oh-my-openagent.
- Use sentence-case headings, straight quotes, and no em dashes.

Plugin README:
- Add "Pick the models": sheet paths, show-sheet.sh, the defaults, and
  why Codex needs its own sheet.
- Link the other-harnesses section and report the 29 of 29 eval result.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The app screens in the film are recreations built from this repository's
real runs and files, and the narration is an AI voice clone. Add a small
caption under the video so viewers know.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@tmcfarlane
tmcfarlane merged commit b890c06 into main Oct 24, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants