Repository navigation
feat(team-avatar): port the remaining workflows as 58 owned skills - #55
Closed
tmcfarlane wants to merge 5 commits into
Closed
tmcfarlane wants to merge 5 commits into
tmcfarlane wants to merge 5 commits into
Conversation
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>
This was referenced Oct 19, 2026
Owner
Author
|
Superseded by #57, the Team Avatar v1.0.0 release, which carries every commit from this PR. The branch is kept. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Requested by Trent
Before: Team Avatar had 15 skills and left out most of pstack: the history and memory skills, the multi-model reviews and bakeoffs, PR babysitting and shipping, the autopilot and orchestration playbooks, the session hook, and model setup.
After: Every pstack skill and playbook has a Team Avatar skill and a character owner. That makes 58 skills across 9 characters. Suki is new, and she leads parallel squads and long autonomous runs. A session hook switches avatar mode on at startup, and
avatar-setuppicks a model per character.Stacked on #54. Merge that one first.
How.
team.jsonrecords each skill's owner and its upstream source.scripts/generate-team-avatar.mjsrebuilds the roster tables inavatar-mode, the plugin README, and every character file from it. The validator now also checks the registry and the generated tables, and it rejects upstream names. Ten writers ported the skills in parallel, one per character. Three independent reviewers then checked fidelity against upstream, Claude Code and Codex compatibility, and the seams between skills, and their findings are fixed. The bun PR watcher and orchestration store live inplugins/team-avatar/tools/with their full test suite. NOTICE.md andlicenses/carry the upstream MIT credits.Verification
Codex 0.159.0 loads all 58 skills, and
$avatar-moderoutes five sample requests to the right character, citing its route table:Claude Code validates and lists version 0.2.0 as enabled. The session hook injects avatar mode, and stays silent with
session hook: off:The validator catches six kinds of injected defects. The tools suite passes 253 of 253, and its typecheck is clean:
End-to-end tests
The same 29 cases ran as real headless sessions in Claude Code (
claude -p) and Codex (codex exec), each in a fresh throwaway repo, graded by the same deterministic checks. Final run: 29 of 29 on both. Full table, every reply, and the history of earlier passes are in docs/evidence/team-avatar-evals.Both runtimes caught the planted SQL injection and the off-by-one in review:
The first pass found one real gap. Claude Code routed to the right skill but did the work without loading it.
avatar-modenow requires loading the routed skill first.Notes for review
watch-pr/transport.test.ts.@ts-expect-errorand empty catch blocks. The type-level refusal checks inwatch-pr/types.compile.tsare now conditional types, and a positive control confirms they still fail on a valid shape. Best-effort cleanup inorch/store.tsuses a named handler.session-start.ps1.effortparameter replaces them.🤖 Generated with Claude Code