Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,22 @@ All notable changes to the claude-plugins project will be documented in this fil

The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). Entries are listed newest-first; each plugin section is treated as released when merged to `main`.

### vibe v1.0.60

#### Changed
- Codex workers default to reasoning effort `high` instead of `xhigh`: new cl-sweep ticket-owner sessions, cl-execute and cl-sweep support lanes, and the vibe skills' native workers. `xhigh` stays available when raised explicitly, and an existing session's persisted effort is never rewritten.
- Worker models follow their role from one policy per harness. In Codex, `cl-sweep/scripts/model-policy.mjs` puts explorer and research lanes on `gpt-5.6-terra`, keeps writers, reviewers and verifiers on `gpt-6-sol`, and leaves mechanical lanes on `gpt-6-luna`; the vibe, handoff and seed-refresh skills take each spawn's model and effort from `vibe/scripts/worker-model.mjs`. In Claude Code, `vibe-backend-worker` moves from opus to sonnet like the other read-only explorers.
- Vibe validation is targeted: the verify worker runs only the tests closedloop-graph's `code_tests_for` returns for each changed file, with Biome, source gates and the affected typecheck. Its full-suite mode is removed, the handoff publication recipe no longer runs `test:affected`, test lanes, `turbo test`, `pnpm verify`, `pnpm test`, `test:lint`, `test:skills`, the Desktop renderer lane or the whole Storybook suite, and Storybook checks run only the new or changed story files. Required browser and Desktop e2e lanes are still recorded for CI evidence.
- Containerized e2e runs only the specs the change needs, one stack per machine.

#### Added
- `vibe/scripts/codex-usage.mjs` reads the Codex account's rate limits through the App Server `account/rateLimits/read` method before workers spawn and when a turn stalls; a reached limit that no longer allows ordinary usage stops the session with a plain message unless the person explicitly allowed credits. A reached limit that still allows ordinary usage continues, with `state` `LIMIT` in the output. With no credits left (`LIMIT-STOPPED`) it stops even when credits were allowed. The read and its classification are codex-fleet-monitor's `ratelimit.mjs` (`readRateLimits` and `classifyUsage`), so the gate and the monitor's `codex-usage` line share one reader and one set of states; the check uses its private stdio App Server transport and never starts the managed daemon.
- The Claude worker launcher reports each turn's last `rate_limit_event` as `rateLimit` and stops a turn that reaches the limit or starts paid extra usage (`USAGE_LIMIT_REACHED`) unless the launch input sets `allowCredits`.
- `vibe/scripts/e2e-lock.mjs`, a machine-wide lock the verify worker runs every containerized e2e command through, taking over a lock left by a process that no longer exists.

#### Removed
- The test-selection budgeting and report admission that only the removed handoff test commands used (`main-sync-budget-ref-read.ts`, the affected test wrapper pin, the `test:lint` budget and the captured-main mapping of `test:affected`).

### code-review v3.12.0

#### Changed
Expand Down
2 changes: 1 addition & 1 deletion plugins/vibe/.claude-plugin/plugin.json
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
{
"name": "vibe",
"description": "Vibe sessions, handoff, seed refresh, and ClosedLoop ticket workflows",
"version": "1.0.59",
"version": "1.0.60",
"dependencies": ["closedloop-core"],
"author": {"name": "ClosedLoop", "email": "support@closedloop.ai"}
}
2 changes: 1 addition & 1 deletion plugins/vibe/.codex-plugin/plugin.json
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
{
"name": "vibe",
"version": "1.0.59",
"version": "1.0.60",
"description": "Vibe sessions, handoff, seed refresh, and ClosedLoop ticket workflows",
"author": {"name": "ClosedLoop", "email": "support@closedloop.ai"},
"homepage": "https://github.com/closedloop-ai/claude-plugins",
Expand Down
27 changes: 23 additions & 4 deletions plugins/vibe/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -183,6 +183,22 @@ and `codex plugin add` for vibe and closedloop-core in Codex) and asks them
to reopen the app. When the check cannot reach GitHub the session carries
on and says so.

Sessions keep Codex and Claude usage down. Each worker's model follows its
role from one policy per harness: in Codex, `scripts/worker-model.mjs` (backed
by `cl-sweep/scripts/model-policy.mjs`, which cl-sweep and cl-execute use too)
puts read-only explorers (requirements, backend, primitive and Storybook
advice, the handoff summary) on `gpt-5.6-terra` and every writer, reviewer
and verifier on `gpt-6-sol`, all at reasoning effort `high` unless raised
explicitly; in Claude Code the agent files' `model:` frontmatter does the
same with sonnet. Before spawning workers or forwarding a request, and when a
turn stalls, Codex runs `scripts/codex-usage.mjs check`, which reads the
account's rate limits from the App Server through codex-fleet-monitor's
shared reader; the Claude launcher reads each turn's `rate_limit_event`. A
reached limit that no longer allows ordinary usage, or paid extra usage in
Claude Code, stops the session with a plain message unless the person explicitly
allowed credits. Containerized e2e runs only the specs the change needs, one
stack per machine through `scripts/e2e-lock.mjs`.

Two notices reach the person without waiting for a reply. When a request names
an outside product as its reference, or gives a screenshot of one, the
requirements worker researches that product's public documentation, reads the
Expand Down Expand Up @@ -249,7 +265,10 @@ request retires, recorded with `retired-tests.mjs` and the retiring ticket or
PRD, and the live ticket lists those deletions next to Missing tests. The
handoff inventory lists repairs and recorded deletions and rejects an added or
unrecorded deleted test file, and the guardrails reviewer checks each one. Handoff
runs the existing affected tests, the writer fixes the session's own code
runs only the targeted existing tests (the ones closedloop-graph's
`code_tests_for` returns for each changed file) with Biome, source gates and
the affected typecheck, never a broad or full local suite; CI runs those. The
writer fixes the session's own code
when one fails, never what a test asserts, and the ticket lists the missing tests as
engineering's scope of work: the behavior and row each covers and the files
to cover. Only the person's explicit request for a pull request authorizes
Expand Down Expand Up @@ -329,11 +348,11 @@ The plugin bundles the ClosedLoop ticket automation skill pack: `cl-policy`, `cl
| `vibe-ticket-worker` | sonnet | Live ticket creation and source link, the session plan as a linked PLN, next-owner lookup, handoff grading, assignment, and cancellation |
| `vibe-environment-worker` | sonnet | Vercel environment creation, redeploy, flag snapshots, Desktop setup, deployment records, and opening a pull request the person asked for at handoff |
| `vibe-change-worker` | sonnet | Sole persistent implementation writer for local planning, full-stack changes, stories, fixes, the handoff missing tests list and serialized recording |
| `vibe-backend-worker` | opus | Read-only decision-table and backend guidance for that writer |
| `vibe-backend-worker` | sonnet | Read-only decision-table and backend guidance for that writer |
| `vibe-primitive-worker` | sonnet | Read-only reusable-component and steward-specification guidance for that writer |
| `vibe-prototype-worker` | sonnet | Read-only canonical prototype guidance or operational immutable Vercel sharing of the writer's reviewed committed result |
| `vibe-storybook-decomposer` | sonnet | Read-only Storybook placement owner (precedent search on sibling entities before any story or shared component is placed) and component extraction, story and control guidance for that writer |
| `vibe-verify-worker` | sonnet | Read-only existing checks, handoff coverage review, full backend suite, Storybook footprint and handoff running app checks with `pnpm control` |
| `vibe-verify-worker` | sonnet | Read-only targeted existing checks (`code_tests_for` per changed file), locked containerized e2e for the specs the change needs, handoff coverage review, Storybook footprint and handoff running app checks with `pnpm control` |
| `vibe-guardrails-reviewer` | sonnet | Read-only judgment-based review of reuse, placement (including Storybook precedent placement), copy provenance, accessibility, and tests |
| `vibe-adversarial-reviewer` | opus | Separate read-only handoff review of ownership, dependencies, contracts, regressions, and web/Desktop behavior |
| `vibe-handoff-summarizer` | sonnet | Read-only plain-language summary of the session inventory, changes, and overlaps |
Expand Down Expand Up @@ -437,7 +456,7 @@ an available update.

## Runtime Files

Version `1.0.58` contains 10 skills and 16 agents. It has no standalone
Version `1.0.60` contains 11 skills and 16 agents. It has no standalone
commands, hooks, root-level shell scripts, or production Python tools under
`tools/python/`; that directory contains the skill-contract test modules and
`vibe_contract_text.py`, the text helpers they share.
Expand Down
2 changes: 1 addition & 1 deletion plugins/vibe/agents/vibe-backend-worker.md
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
---
name: vibe-backend-worker
description: Read-only backend advisor for the single persistent vibe implementation writer. Checks route/service ownership, shared types, auth/org validation, schema/migration and seed requirements. Its implementation guidance is applied by the sole writer, not another backend author.
model: opus
model: sonnet
tools: Read, Grep, Glob, Bash, Skill
---

Expand Down
17 changes: 9 additions & 8 deletions plugins/vibe/agents/vibe-environment-worker.md
Original file line number Diff line number Diff line change
Expand Up @@ -214,14 +214,15 @@ every conflict/source problem. Do not run a raw merge/commit/push yourself.
exemption: preserve dirty code/records and BLOCK through the same-writer
workaround/exclusion/restore flow, never stash/copy/delete them.
A build-loop deploy runs only what the commit hooks run: the affected
typecheck and Biome on the changed source, no test suites. At handoff only:
Run the existing affected tests before pushing through that SAME writer's
executor, from the worktree:
`TURBO_CONCURRENCY=2 pnpm turbo test --filter="...[<since>]" --continue`,
where `<since>` is the session's `vercel.lastDeployedCommit`, or its
`baseCommit` before the first deploy. That is the tests of every package
the session changed since then plus the packages that depend on them
(Storybook's story sweep included). The shared executor owns fixed selected
typecheck and Biome on the changed source, no test suites. At handoff the
matrix adds the source gates, the Storybook catalog check when stories
changed, and the typecheck of every package the session changed since
`<since>` plus the packages that depend on them
(`pnpm turbo typecheck --filter="...[<since>]"`, where `<since>` is the
session's `vercel.lastDeployedCommit`, or its `baseCommit` before the
first deploy), still no broad or full test suite (operator ruling,
2026-10-19): the verify worker runs the tests `code_tests_for` selects for
each changed file, and CI runs the broad suites. The shared executor owns fixed selected
process budgets and the Root-bound validation envelope; never supply a
timeout override or change any test/assertion/hook deadline. A failure the
brief lists as a deliberately obsolete expectation must carry the exact
Expand Down
4 changes: 3 additions & 1 deletion plugins/vibe/agents/vibe-primitive-worker.md
Original file line number Diff line number Diff line change
Expand Up @@ -53,7 +53,9 @@ skill (`.claude/skills/storybook/SKILL.md`, `references/gotchas.md`,
one lifecycle tag and the release bump, never a frozen v1.0 root
(`guardrails.md`, "Storybook taxonomy").
4. `pnpm --filter storybook catalog:sync`, `pnpm --filter storybook validate:catalog`,
`pnpm --filter storybook test`, and Biome on your files, until all pass.
`pnpm --filter storybook exec vitest run <the new story files>` (those
files only, never the whole Storybook suite), and Biome on your files,
until all pass.
5. Append to the session change log (`$(git -C <wt> rev-parse --absolute-git-dir)/vibe-changes.md`).
6. Add a Progress line to the live ticket per `ticket-template.md` naming the
new building block and its story.
Expand Down
3 changes: 2 additions & 1 deletion plugins/vibe/agents/vibe-storybook-decomposer.md
Original file line number Diff line number Diff line change
Expand Up @@ -193,7 +193,8 @@ Placement:

As an advisor, run no writing commands. Inspect the current result and report
what the sole writer should check: `pnpm exec biome check` and
`pnpm --filter storybook test` for the changed stories. Report: files
`pnpm --filter storybook exec vitest run <changed story files>` for the
changed stories only, never the whole Storybook suite. Report: files
extracted, stories added or updated (path and story names), each story or
shared component placed differently from the recorded Placement block or
without one, and any component not yet covered with the reason. Do not claim
Expand Down
39 changes: 25 additions & 14 deletions plugins/vibe/agents/vibe-verify-worker.md
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ worktree/branch.

## Inputs and references

The owned worktree, mode (`checks`, `footprint`, `full-suite`, `coverage`, or
The owned worktree, mode (`checks`, `footprint`, `coverage`, or
`control`), inventory, acceptance criteria and the sole writer's local plan and
missing tests list.
Read the actual table at the same session decision table path and affected/
Expand Down Expand Up @@ -58,12 +58,22 @@ automatic technical approval question for the vibe coder. Verify real canonical
run/attempt/job/step and checked-out source evidence, not run.head_sha or a
generic log SHA; missing/unsupported evidence remains a precise final gap.

Run Biome WITHOUT write/fix flags, source gates, `pnpm typecheck:affected`,
`pnpm test:affected --continue`, and the relevant existing lanes named by
`pnpm test:lanes`. Shared app changes also run the Desktop renderer lane
directly when the repo requires it. Graph coverage identifies missing existing
suites, not permission to write one. Three more, from the reviews of
symphony-alpha #8885 and #8886 (operator ruling, 2026-10-16):
Validation is targeted, never a broad or full local suite (operator ruling,
2026-10-19, after vibe sessions exhausted the weekly Codex limits). Run Biome
WITHOUT write/fix flags on the changed files, source gates,
`pnpm typecheck:affected` (the changed packages and their dependents), and
only the tests closedloop-graph's `code_tests_for` returns for each changed
file, each through its owning package's runner on those test files alone (for
example `pnpm --filter <package> exec vitest run <test files>`). Never run
`pnpm test`, `pnpm verify`, `pnpm test:affected`, `pnpm test:lanes --exec`, a
whole lane or a package's whole suite locally; CI runs the broad suites on the
pull request and after merge, and the repo's own pre-push hook still runs
`pnpm verify --scope=push`. A requested pull request opens once these targeted
checks pass. When `code_tests_for` answers `found: false` for a changed file,
record that file and run no wider suite in its place. Graph coverage
identifies missing existing suites, not permission to write one. Three more,
from the reviews of symphony-alpha #8885 and #8886 (operator ruling,
2026-10-16):

- When the diff touches Storybook sidebar order (`apps/storybook/story-group-order.ts`,
`apps/storybook/.storybook/preview.tsx` or `manager.ts`) or adds, moves or
Expand All @@ -74,7 +84,14 @@ symphony-alpha #8885 and #8886 (operator ruling, 2026-10-16):
- Every e2e spec the session touched, and every spec that covers a touched
screen (`code_tests_for`, and `rg` of the screen's route and test IDs under
`e2e/`), runs at handoff through the repo's containerized path (the
`local-ci` skill, `dagger call e-2-e`), never natively on the person's Mac.
`local-ci` skill, `dagger call e-2-e`), never natively on the person's Mac,
and only those specs, never the whole e2e suite. The containerized stack
is one per machine (one compose project name, fixed ports and shared
Dagger cache volumes), so start it only through
`node ../skills/vibe/scripts/e2e-lock.mjs run <the e2e command>`, which
waits until no other session's stack is running. Its exit 75 means the
stack stayed busy past the wait: record that limitation with the holder it
prints, never a pass, and never start the stack around the lock.
When Docker or Dagger is missing, record that exact limitation; never claim
the spec passed.
- Before you claim that CI runs, or ran, a check or test, read
Expand Down Expand Up @@ -107,12 +124,6 @@ Do not change code, suppress a check, shrink an allowlist yourself, skip an
assertion, raise a timeout or update a snapshot to clear it. List pre-existing
failures accurately. Existing tests run; nobody writes them in vibe or handoff.

## Full-suite mode

When backend changed, run `pnpm verify` unscoped, `pnpm test`, and relevant
script lanes (`pnpm test:lint`, `pnpm test:skills` where named). Same read-only
and displayless rules. Return pass/fail per lane; the writer fixes defects.

## Coverage advice and the missing tests list

Report needed new coverage into the local plan, never author it. Check source
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -80,9 +80,9 @@ SUBAGENTS
so route anything needing closedloop-graph to a worker that has them, and
tell workers to load the tools with ToolSearch first.
- Codex: use same-family `spawn_agent` workers with `fork_turns: "none"`
and `model: "gpt-6-sol"` at `high` or `xhigh`. They inherit no
conversation, so give each one the plan path, these sources, and its
question.
and `model: "gpt-6-sol"` at `high` (`xhigh` only when you raise it
explicitly for that worker). They inherit no conversation, so give each
one the plan path, these sources, and its question.
- When you run as a support lane, which never delegates, do the workers'
reads yourself.
- You own the conclusions. Verify load-bearing worker claims against primary
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -82,14 +82,19 @@ and repo-doc constraints discovered by the worker. Do not rely on conversation
inheritance because `fork_turns: "none"` supplies none.

Every support `spawn_agent` packet must set a role-bounded model and
`reasoning_effort`. Use `model: "gpt-6-sol"` with `high` or `xhigh` for
planning review, correctness-sensitive review/QA, synthesis, bounded
exploratory code/evidence analysis, and other high-judgment work where
mistakes matter. Only when the ticket owner actually decomposes a separate
bounded, deterministic lower-risk support role may it use
`model: "gpt-6-luna"` with `low` for deterministic inventory, artifact
parsing, hashing, or status mechanics. Luna returns evidence only; it never
owns orchestration, ticket work, coding/planning, final judgment, or mutation.
`reasoning_effort`. The bindings come from the plugin's one Codex model
policy, `../../cl-sweep/scripts/model-policy.mjs`. Use `model: "gpt-6-sol"`
with `reasoning_effort: "high"` for planning review, correctness-sensitive
review/QA, synthesis, and other high-judgment work where mistakes matter;
raise it to `xhigh` only by an explicit choice for that lane. Use
`model: "gpt-5.6-terra"` with `high` for explorer lanes: bounded exploratory
code/evidence analysis, codebase exploration, graph lookups, requirements or
precedent research, and read-only search. Only when the ticket owner actually
decomposes a separate bounded, deterministic lower-risk support role may it
use `model: "gpt-6-luna"` with `low` for deterministic inventory, artifact
parsing, hashing, or status mechanics. Luna and terra return evidence only;
they never own orchestration, ticket work, coding/planning, final judgment,
or mutation.
These are explicit role/model choices, not fallback models. The model names
here are Codex models for a Codex worker. A Claude Code worker launches the
same lanes as its own sub-agents on Claude models and never routes a lane
Expand All @@ -100,10 +105,10 @@ Model or effort selection alone is not a token-savings claim.

For every path-addressed read-only planning, code-review lens/synthesis,
logical-QA, visual-QA, audit, search, or status support spawn, use
`fork_turns: "none"` plus the role-appropriate `gpt-6-sol` or `gpt-6-luna`
selection above and its matching `reasoning_effort`. If the work has not been
decomposed into a genuinely bounded deterministic lower-risk support role, use
`gpt-6-sol`.
`fork_turns: "none"` plus the role-appropriate `gpt-6-sol`, `gpt-5.6-terra`
or `gpt-6-luna` selection above and its matching `reasoning_effort`. If the
work is neither an explorer lane nor a genuinely bounded deterministic
lower-risk support role, use `gpt-6-sol`.

The self-contained prompt must state the exact absolute input artifact path and
SHA-256, cwd, requirements reference, expected absolute result path, and bounded
Expand Down
Loading
Loading