@brainervirus/workit-claude-code 7.6.0 → 7.7.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@brainervirus/workit-claude-code",
-  "version": "7.6.0",
+  "version": "7.7.0",
   "private": false,
   "description": "Workit Claude Code plugin — session and per-turn task context, branch policy on git shell commands, workit method skills, and verifier/reviewer/implementer agents",
   "keywords": [
@@ -39,8 +39,8 @@
     "build": "bun scripts/build.ts"
   },
   "devDependencies": {
-    "@brainervirus/workit-cli": "^7.6.0",
-    "@brainervirus/workit-core": "^7.6.0"
+    "@brainervirus/workit-cli": "^7.7.0",
+    "@brainervirus/workit-core": "^7.7.0"
   },
   "engines": {
     "node": ">=24"

package/skills/fanout/SKILL.md

@@ -8,60 +8,60 @@ description: Run independent slices in parallel - one worker per isolated worktr
 Fan out only independent slices: disjoint files, no shared mutable state, each
 verifiable alone. Code-coupled work stays with one owner, who fans out after
 the blocking part lands. A worker whose whole job is re-running one command is
-ceremony; do it yourself.
+ceremony; do it yourself. Workit never spawns agents; you do.
 
 1. **Plan the slices** (call the Skill tool with `workit:shape`) in a plan file (`references/brief.md`):
-   per slice an id, branch, TIER, the file-scope manifest (SCOPE globs),
-   `owns` for shared files (lockfile, registry, barrels), `dependsOn` only for
-   a real dependency (independent PRs off trunk are the default), and the
-   brief fields. `workit fanout plan <plan.json>` refuses an empty or
-   placeholder (`<goal>`, `TBD`) brief field (exit 2) and two slices that may
-   write one file (exit 3) with a fix: an owner for a shared file, or a
-   dependency that serializes them. Apply it and re-run: refuse to spawn while
-   a field is empty or the plan is refused. Its `waves` say which slices may
-   run together; keep 4-6 in flight.
-2. **Brief each worker** from its slice with the fixed template: GOAL, SCOPE,
-   CONTEXT (pointers, not pasted text), ACCEPTANCE (Given/When/Then), VERIFY
-   (exact commands), TIER, TIMEBOX, SCRATCH, FORBIDDEN, REPORT, STANDING.
-   STANDING is every standing order and user directive so far, pasted
-   verbatim into each spawn and respawn: directives decay across resumes.
-3. **Spawn all workers in one message**, in the background, each in its own
-   worktree: Claude Code's `implementer` agent (first command `workit git
-   branch <branch> --base <base>`); elsewhere `workit fanout worktree create
-   <slice>`, which prints the SCRATCH dir. Refill from `fanout status`.
-4. **Judge liveness by side effects only:** `workit fanout status` (head age,
-   PR, CI, verdict, landed; STUCK past the TIMEBOX, default 30 min). Stop a
-   stuck worker and observe that it exited (a timeout is not proof). Removing
-   a worktree drops its uncommitted changes: `workit fanout worktree release
-   <slice>` records `git status` in the ledger first and refuses them without
-   `--force` (native worktrees: record `git -C <wt> status --short` first).
-   Respawn in `MODE: resume` (brief, directives, last report), same branch.
-   Never two live workers on one branch. Replace at most twice, then re-slice
-   or report the gap. Never chain resumes.
-5. **Verify each slice independently.** A fresh agent that did not write it
-   (Claude Code: the `verifier` agent) runs VERIFY and verify-<app>, then
-   `workit ledger verdict <result> --branch <b> --how "<evidence>"` under the
-   session you started it with (`WORKIT_SESSION_ID=<lead>-v<n>`, set by you,
-   never chosen by the author; Claude Code: the hook names one).
-   A worker's report is a pointer, never evidence.
-6. **Fan in** with `workit fanout check`: out-of-scope files (any file outside
-   it stops the fan-in), and `git merge-tree` conflicts with trunk and between
-   siblings, charged to the slice that lands later. Fix what it names (an
-   out-of-scope edit becomes a follow-up slice) until it exits 0. A slice
-   whose PR merged reads as landed; its dependents stop waiting. Then
-   `workit ledger check --branch <b>` per slice, land in `fanout status` order
-   and release the worktrees you made. Stacks: `workit stack plan <bottom> …
-   <top>` once, then `stack sync` and `land`. Only you touch topology: workers
-   never rebase, retarget or merge. Then ship (call the Skill tool with `workit:ship`).
+   per slice an id, branch, TIER, SCOPE globs, `owns` for shared files,
+   `dependsOn` only for a real dependency, and the brief fields. `workit fanout
+   plan <plan.json>` refuses an empty or placeholder brief field (exit 2) and
+   two slices that may write one file (exit 3) with a fix; apply it and re-run:
+   refuse to spawn while a field is empty or the plan is refused. One PR per
+   slice is the default; `"fanIn": "integration"` only when the user wants one
+   PR (`references/orchestration.md`).
+2. **Record standing orders** (`workit ledger standing add "<order>"`): every
+   user directive all workers share. Directives decay across resumes.
+3. **Brief each worker verbatim** with `workit fanout brief <slice>`: plan
+   fields, standing orders, SCRATCH, session id, fan-in rule. Never hand-edit
+   it; change the plan or the standing orders and render again.
+4. **Spawn in a rolling window**, in the background, each in its own worktree:
+   keep 4-6 in flight and refill from `spawnable` in `workit fanout status` as
+   each one finishes. Pick the model from TIER (`references/orchestration.md`).
+   Claude Code: the `implementer` agent; elsewhere first `workit fanout
+   worktree create <slice>`.
+5. **Judge liveness by side effects only:** `workit fanout status` (head age,
+   PR, CI, verdict, landed; STUCK past the TIMEBOX). Stop a stuck worker and
+   observe that it exited (a timeout is not proof). Removing a worktree drops
+   its uncommitted changes: `workit fanout worktree release <slice>` records
+   `git status` in the ledger first and refuses them without `--force`. Never
+   two live workers on one branch.
+6. **Retry once, then escalate.** A stuck or failed slice gets one retry with
+   a fresh brief (`workit fanout brief <slice> --mode resume`, same branch);
+   if that fails too, re-slice it, take it over, or report the gap. Never
+   chain resumes. A hard slice may race instead: N attempts, keep the best
+   verdict (`references/orchestration.md`).
+7. **Verify by the workspace `verification` setting** (`workit grant show`).
+   A report is a pointer, never evidence. `self`: a verifier that wrote none
+   of the slices, or you if you wrote none. `independent` or high risk: a
+   separate verifier session, never yours (doctrine: the ledger only refuses
+   authors). One verifier may take a batch of slices, one verdict per branch
+   (`workit ledger verdict <result> --branch <b> --how "<evidence>"`), in a
+   session never chosen by the author. A review panel on separate models only at high risk.
+8. **Fan in** with `workit fanout check`: out-of-scope files (any file outside
+   it stops the fan-in) and `git merge-tree` conflicts with trunk and between
+   siblings. Fix what it names until it exits 0; a merged slice reads as
+   landed. Then `workit ledger check --branch <b>` per slice, land in `fanout
+   status` order, release the worktrees you made, and `workit ledger standing
+   clear`. Only you touch topology: workers never rebase, retarget or merge,
+   except the integration-tip merge in integration mode. Then ship
+   (call the Skill tool with `workit:ship`).
 
 ## Example
 
 Bad brief: "Do the API part and add tests." (no scope, no acceptance, no
 verify command, so nobody can tell when it is done)
 
-Good brief: `references/brief.md` (GOAL: `GET /v1/usage` returns daily run
-counts; SCOPE: `src/routes/usage.ts`, `test/usage.test.ts`; VERIFY:
-`workit check test`; ...).
+Good brief: the output of `workit fanout brief usage-endpoint`
+(`references/brief.md` shows one).
 
 ## Check
 

package/skills/fanout/references/brief.md

@@ -1,8 +1,11 @@
 # Worker brief template
 
-Every field is required. A brief with an empty field is not spawned. Point to
-files and ledger rows instead of pasting their content. Size the slice so the
-worker finishes it in about 150k tokens of context; split a larger one first.
+`workit fanout brief <slice>` renders this brief from the plan, the standing
+orders in force (`workit ledger standing list`) and the slice's scratch dir;
+pass its output verbatim. The template below is what it fills in. Every field
+is required. A brief with an empty field is not spawned. Point to files and
+ledger rows instead of pasting their content. Size the slice so the worker
+finishes it in about 150k tokens of context; split a larger one first.
 
 ```md
 MODE: <new | resume (a replacement continuing an existing branch)>
@@ -16,11 +19,12 @@ VERIFY: <exact commands, e.g. `workit check test`, the verify-<app> feature to d
 TIER: <mundane | standard | hard: how much model the slice needs>
 TIMEBOX: <wall clock or turn budget; past it without a new commit you will be replaced>
 SCRATCH: <your own temp dir: the one `fanout worktree create` printed, else `mktemp -d`; never a shared path>
-FORBIDDEN: <no edits outside SCOPE; no rebase, retarget, merge or force-push; no new dependencies; ...>
+FORBIDDEN: <no edits outside SCOPE; no new dependencies; ...>
+FAN-IN: <one PR per slice: no rebase, retarget, merge or force-push | integration: merge the integration tip before reporting, nothing else>
 REPORT: branch, head SHA, files changed, each VERIFY command with its exit code,
   each ACCEPTANCE line met / not met, rulings you made (`workit ledger ruling`),
   anything out of scope as a follow-up, not a diff.
-STANDING: <the standing orders, verbatim: user preferences and every directive given so far>
+STANDING: <the standing orders, verbatim: `workit ledger standing add` records each one>
   export WORKIT_SESSION_ID=<lead>-w<n>   (set by the lead; a verifier brief gets <lead>-v<n>)
 ```
 
@@ -40,7 +44,8 @@ VERIFY: `workit check test`; verify-api feature "usage"
 TIER: standard
 TIMEBOX: 45 minutes
 SCRATCH: ../app-wt/usage-endpoint/.workit-scratch
-FORBIDDEN: no edits outside SCOPE; no schema migration; no rebase or force-push; no new packages
+FORBIDDEN: no edits outside SCOPE; no schema migration; no new packages
+FAN-IN: one PR per slice. Never rebase, retarget, merge or force-push: the lead owns topology.
 REPORT: as in the template
 STANDING: conventional commits; no comments that restate code; ask nothing, record rulings instead
 ```
@@ -55,7 +60,9 @@ stack); `worktree` defaults to `../<repo>-wt/<id>`. `owns` claims a shared
 file another slice's glob also matches. Globs that match no file yet are compared
 through a sample path. Use `/` as the separator; a backslash only escapes
 literal brackets: `"app/\\[id\\]/page.tsx"`. A `timebox` such as `45 minutes`
-is the slice's STUCK threshold in `workit fanout status`.
+is the slice's STUCK threshold in `workit fanout status`. `"fanIn":
+"integration"` (with an integration branch as `trunk`) is the one-PR mode in
+`orchestration.md`; the default `"prs"` lands one PR per slice.
 
 ```json
 {
@@ -90,7 +97,7 @@ is the slice's STUCK threshold in `workit fanout status`.
 }
 ```
 
-## Worker rules (paste into the brief when the host has no implementer agent)
+## Worker rules (`workit fanout brief` appends them as RULES)
 
 1. First command. `MODE: new`: `workit git branch <branch> --base <base>` (the
    worktree may start on a name that breaks branch policy). `MODE: resume`:

package/skills/fanout/references/orchestration.md

@@ -0,0 +1,117 @@
+# Running a fanout
+
+How the lead picks models, keeps workers flowing, retries, races, verifies
+and fans in. Workit records and checks; the host's own subagents run the work.
+
+## TIER to model
+
+TIER is a field of each slice and of its brief. Map it where the host lets you
+pick a model per spawn; nothing in the workspace config routes it.
+
+| TIER | Claude Code (`model` on the Agent call) | Hosts without a per-spawn model |
+| --- | --- | --- |
+| scouting (read-only search, not a slice) | `haiku` | host default |
+| mundane | `sonnet` | host default |
+| standard | omit it (inherits yours) | host default |
+| hard | omit it (inherits yours); consider a race | host default |
+
+OpenCode takes a subagent's model from its agent config: pick a configured
+agent whose model fits the TIER, else the default. Codex, Cursor and Pi run
+every worker on the host's model, so TIER only sizes the timebox and decides
+whether to race.
+
+## Rolling window
+
+Keep 4-6 workers in flight on Claude Code; elsewhere as many as the host runs
+in the background, never more than 6. Do not wait for a whole wave: when one
+worker finishes (its notification, or a new head or PR in `workit fanout
+status`), verify it, then spawn the next slice from `spawnable` in the same
+status output. A slice whose dependencies landed, or a stacked child whose
+parent is verified, appears there by itself.
+
+## Retry, then escalate
+
+A slice is retried when it is STUCK, its worker exits without meeting
+ACCEPTANCE, or its verdict fails. Stop the worker and observe that it exited.
+Release its worktree: `workit fanout worktree release <slice>` (native
+worktrees: record `git -C <wt> status --short` first). Then spawn one fresh
+worker with `workit fanout brief <slice> --mode resume`: same branch, and the
+brief points at its head and its ledger rows (last report, verdicts,
+rulings). If the retry fails too, escalate instead of a third worker:
+re-slice it smaller (`workit fanout plan` again), take it over yourself, or
+report the gap to the user.
+
+## Race a hard slice
+
+When a hard slice has an uncertain approach and a cheap VERIFY, run 2-3
+attempts at once instead of retrying in series. `workit fanout brief <slice>
+--attempt <n>` gives each attempt its own branch (`<branch>-try<n>`), session
+and scratch dir. Each attempt needs its own worktree: Claude Code's
+`implementer` isolates natively; on other hosts `fanout worktree create` makes
+only the slice's own worktree, so run the attempts one after another there.
+Verify every attempt, keep the one with the best verdict (accepted first,
+then the fewest changed files, then the earliest), point the slice branch at
+it (`git branch <branch> <winner>`), have the verifier record the verdict on
+`<branch>` (same head), and delete the losing attempt branches.
+
+## Verification
+
+Follow the workspace `verification` setting (`workit grant show`); the
+worker never records a verdict on its own work.
+
+- `self` (the default): a verifier that wrote none of the slices records the
+  verdicts. A lead that authored none of the slices may record them in its
+  own session.
+- `independent`, and any slice at high risk: a separate verifier session
+  records them, never the lead's own; at high risk one verifier per slice,
+  plus a review panel (below). This is doctrine, not enforced: the ledger
+  only tells a branch's authors from everyone else, and a lead that wrote
+  none of it counts as independent.
+
+Which session id a verdict carries, per host:
+
+- Claude Code: subagents inherit your `WORKIT_SESSION_ID`, so the
+  SubagentStart hook names each `verifier` and `reviewer` its own session
+  (`<lead>:<agent id>`), passed as `--session`; that one wins over any
+  `<lead>-v<n>` in a brief. An `implementer` uses the session its brief
+  exports for its workit commands.
+- Other hosts: the session the brief sets wins: `export
+  WORKIT_SESSION_ID=<lead>-w-<slice>` from `workit fanout brief` for a
+  worker, `<lead>-v<n>` (or `--as verifier`) for a verifier you start.
+
+A batch verifier takes several slices on one surface in one session: for
+each branch it checks out the head, runs that slice's VERIFY, and records its
+own `workit ledger verdict <result> --branch <b> --how "<evidence>"`. Every
+verdict is keyed to that branch's head SHA; a slice whose head moves after
+it needs a new one. Batch at most what fits one context, about 4 slices.
+
+## Review panel (high risk only)
+
+At risk=high only, 2-3 fresh reviewers (call the Skill tool with `workit:review`), each on a model
+other than the author's (Claude Code: a different `model`), each records
+`workit ledger verdict <result> --kind review --branch <b> --how "<findings>"`.
+Land only when every panel verdict passes. Below high risk a panel costs
+more than it finds.
+
+## Integration-branch fan-in (one PR)
+
+Only when the user wants one PR for the whole fanout:
+
+1. Cut the integration branch (`workit git branch <integration> --base
+   <trunk>`) and plan with `"fanIn": "integration"` and `--trunk
+   <integration>`. Planning refuses integration mode on origin's default
+   branch.
+2. Every brief then tells the worker to merge the integration tip into its
+   branch before reporting (the fast-forward rule). That merge is the only
+   one a worker makes, and only in this mode.
+3. Land each verified slice with `git merge --ff-only <slice-branch>` in the
+   integration checkout. Not a fast-forward (the tip moved since the worker
+   merged)? Resume the worker to merge again; never a merge commit of your own.
+4. `workit fanout check` gates against the integration branch; a slice whose
+   tip is on it reads as landed. When all have landed, ship the integration
+   branch as one PR (call the Skill tool with `workit:ship`).
+
+## Stacks
+
+For stacked slices (`dependsOn` one parent): `workit stack plan <bottom> …
+<top>` once, then `workit stack sync` and `workit stack land`.