opencode-swarm 7.110.2 → 7.111.0

package/.opencode/skills/ci-failure-batching/SKILL.md

@@ -6,7 +6,8 @@ description: Batch collection and fix protocol for CI failures. Triggered when a
 # CI Failure Batching
 
 ## Trigger
-When ANY CI check fails on the PR (pr-monitor surfaces `pr.ci.failed`).
+When the PR monitor surfaces `pr.ci.failed`. The event is batched after the
+check set is complete and includes all known failed checks in `failedChecks`.
 
 ## Protocol
 1. **DO NOT immediately fix the first failure.** Check if other jobs are still running:
@@ -29,11 +30,7 @@ Example from session #1685:
 - Without batching: 6 pushes (format → stale-assertion-1 → stale-assertion-2 → integration → merge-group → clean)
 - With batching: 2 pushes (collect all → fix all → push once → clean)
 
-## Pr-monitor workaround
-The pr-monitor fires `pr.ci.failed` per-check. When the first event arrives:
-1. Check `gh pr checks` — are other jobs still running?
-2. If yes: WAIT for completion
-3. If no: the single failure IS the only failure — proceed normally
-
-## Root cause
-The pr-monitor should batch-fire after the CI run completes (issue #1746 item 7). This playbook is the manual protocol.
+## Pr-monitor expectation
+The pr-monitor should fire one `pr.ci.failed` event for the completed failing
+check set, not one event per check. Still verify with `gh pr checks` before
+fixing, because GitHub can append late merge-group or matrix jobs.

package/.opencode/skills/gate-attribution/SKILL.md

@@ -1,24 +1,38 @@
 ---
 name: gate-attribution
-description: Per-task gate dispatch protocol. Documents the single-taskId attribution rule and parallel-lane optimization for reviewer/test_engineer gates.
+description: Per-task gate dispatch protocol. Documents single-task attribution plus parseable set-dispatch reviewer/test_engineer rows.
 ---
 
 # Gate Attribution
 
 ## The rule
-The gate tracker attributes reviewer/test_engineer dispatches PER TASK (single taskId in the prompt). Set-dispatches covering multiple tasks do NOT count per-task, even with per-task verdicts.
+The gate tracker attributes reviewer/test_engineer dispatches PER TASK. A
+single-task prompt still attributes by `task_id` / `taskId` / unambiguous prompt
+task ID. A set-dispatch can also count per-task when the reviewer/test_engineer
+output includes parseable per-task rows:
+
+```
+[REVIEWED] | task-2.1 | APPROVED | ...
+[REVIEWED] | 2.2 | PASS | ...
+```
+
+Rows with `task-X.Y` are normalized to `X.Y`; unsafe or non-plan IDs are ignored.
+Only passing verdicts (`APPROVED`, `PASS`, or `PASSED`) create gate evidence.
+Rows such as `NEEDS_REVISION` are reviewed rows but do not satisfy the gate. If
+no reviewed rows are parseable, attribution falls back to the single-task rule.
 
 ## Protocol
-1. **For each task requiring a gate:** Dispatch a separate reviewer and/or test_engineer with exactly ONE taskId
-2. **Minimize overhead via parallel dispatch:**
+1. **For unrelated or high-risk tasks:** Dispatch separate reviewer and/or test_engineer lanes with exactly ONE taskId.
+2. **For a true set-dispatch:** Require one `[REVIEWED] | task-id | verdict | ...` row per task in the returned output. Only passing verdict rows count.
+3. **Minimize overhead via parallel dispatch when set-dispatch is not appropriate:**
    ```
    dispatch_lanes_async with:
    - common_prompt: shared verification context
    - lanes: one lane per task, each with a single taskId
    - max_concurrent: up to 3
    ```
-3. **Collect + attribute:** Each lane result auto-attributes to its taskId
-4. **Do NOT batch:** Even for identical reviews, dispatch separately
+4. **Collect + attribute:** Single-task lanes auto-attribute to their taskId; set-dispatch rows auto-attribute per parsed row.
+5. **Do NOT rely on prose summaries:** A batched dispatch without parseable rows is ambiguous and does not count per-task.
 
 ## Optimization for trivial tasks
 For pure ceremony gates (1-line doc fix):
@@ -28,4 +42,7 @@ taskId: X.Y
 ```
 
 ## Why this exists
-The gate tracker (`src/hooks/delegation-gate.ts`) keys delegation chains by `sessionID`, and task attribution resolves exactly ONE taskId per dispatch (`args.task_id ?? args.taskId`). Ambiguous multi-task prompts fail closed (resolve to null) rather than attributing to any task — so a set-dispatch covering multiple tasks does not count per-task even with per-task verdicts. Tracked in issue #1746 item 6.
+The gate tracker (`src/hooks/delegation-gate.ts`) keys delegation chains by
+`sessionID`. Ambiguous multi-task prompts still fail closed, but parseable
+`[REVIEWED] | task-id | ...` rows provide explicit per-task attribution for
+set-dispatches. Tracked in issue #1746 item 6.

package/.opencode/skills/merge-queue-readiness/SKILL.md

@@ -10,21 +10,33 @@ Before adding the PR to the merge queue (or before the final push if the repo us
 
 ## Protocol
 1. **Fetch latest main:** `git fetch origin main`
-2. **Create a temporary simulation worktree (do NOT mutate the PR branch).** Use a project-relative path UNDER the swarm worktree base so the path is portable across OSes and its later removal is permitted by the worktree guardrail (paths outside `.swarm-worktrees/` are blocked). Do NOT hardcode `/tmp` — it does not exist on Windows.
+2. **Run the simulation command (preferred):**
+   ```
+   /swarm ci-simulate --base origin/main --head <pr-branch>
+   ```
+   By default this runs fixed local CI gates: `bun run typecheck`,
+   `bun run lint:ci`, `bun run build`, `bun run test:unit:ci`,
+   integration/security/smoke tests, and `bun run drift:check`.
+   `test:unit:ci` runs unit tests per file with the repo's quarantine filters
+   and retry budget, matching CI's mock-isolation semantics. The command creates
+   a temporary detached worktree under `.swarm/ci-simulate`, merges the PR branch
+   or current worktree commit, runs the gate sequence, removes the worktree, and
+   prunes metadata. It does not accept arbitrary shell commands.
+3. **Manual fallback:** If the command is unavailable, create a temporary simulation worktree (do NOT mutate the PR branch). Use a project-relative path UNDER the swarm worktree base so the path is portable across OSes and its later removal is permitted by the worktree guardrail (paths outside `.swarm-worktrees/` are blocked). Do NOT hardcode `/tmp` — it does not exist on Windows.
    ```
    git worktree add .swarm-worktrees/merge-sim origin/main
    cd .swarm-worktrees/merge-sim
    git merge <pr-branch> --no-edit
    ```
-3. **Run integration + unit tests against the merged result:**
+4. **Run integration + unit tests against the merged result:**
    ```
    bun test tests/integration --timeout 120000
    bun test tests/unit --timeout 120000
    ```
    (Use per-file loops for hot modules per AGENTS.md invariant 6)
-4. **If failures:** Fix on the PR branch, re-push, re-simulate. Always run the cleanup step (5) before re-simulating or on any exit path — do not leave the simulation worktree behind.
-5. **Cleanup (run on EVERY exit path, including failure):** `git worktree remove --force .swarm-worktrees/merge-sim`. If the remove is guardrail-blocked or fails, delete the directory directly and run `git worktree prune`.
-6. **Only after simulation passes,** add PR to the merge queue.
+5. **If failures:** Fix on the PR branch, re-push, re-simulate. Always run the cleanup step (6) before re-simulating or on any exit path — do not leave the simulation worktree behind.
+6. **Cleanup for manual fallback (run on EVERY exit path, including failure):** `git worktree remove --force .swarm-worktrees/merge-sim`. If the remove is guardrail-blocked or fails, delete the directory directly and run `git worktree prune`.
+7. **Only after simulation passes,** add PR to the merge queue.
 
 ## Why this matters
 PR-branch CI and merge-group CI test DIFFERENT things:
@@ -33,5 +45,7 @@ PR-branch CI and merge-group CI test DIFFERENT things:
 
 Integration tests that pass on the PR branch may fail in merge-group context due to test interactions exposed by the merged result.
 
-## Root cause
-A `swarm ci-simulate` command would automate this (issue #1746 item 5). This playbook is the manual protocol.
+## Automation
+`/swarm ci-simulate` automates the merge-result worktree, merge, command run,
+worktree removal, and metadata prune. Use the manual protocol only when the
+command is unavailable or the repo needs bespoke setup.

package/.opencode/skills/skill-edit-validation/SKILL.md

@@ -19,11 +19,18 @@ After editing ANY `.md` file under `.opencode/skills/`, `.claude/skills/`, or `.
    - Does the assertion still hold against the new content?
    - Is it checking a substring containing the old phrase?
    - Is it checking for the ABSENCE of a word the new wording introduces? (e.g., `not.toContain('skip')` catches "this check is skipped")
-4. **Update stale assertions in the same changeset.** Do NOT defer to CI.
-5. **Preserve behavioral intent:** When updating, preserve what the assertion TESTS (e.g., "the plan skill has a spec-absent branch"), not just the string match.
+4. **Prefer the semantic registry:** If the assertion is checking skill behavior
+   rather than an exact contract string, move it behind
+   `tests/helpers/skill-content-registry.ts` (or add a concept there) and assert
+   the named concept from the test.
+5. **Update stale assertions in the same changeset.** Do NOT defer to CI.
+6. **Preserve behavioral intent:** When updating, preserve what the assertion TESTS (e.g., "the plan skill has a spec-absent branch"), not just the string match.
 
 ## Constraint
 Do NOT rubber-stamp brittle assertions. If an assertion tests implementation detail rather than behavioral intent, flag it for refactoring to a semantic check.
 
 ## Root cause
-A skill-content test registry would auto-adjust (issue #1746 item 3). This playbook is the manual safety net.
+Skill-content tests should assert named semantic concepts where possible. The
+registry in `tests/helpers/skill-content-registry.ts` is the preferred safety
+net for recurring skill wording checks; use the manual grep sweep for exact
+contract strings and any tests not yet migrated.

package/.opencode/skills/worktree-retry-cleanup/SKILL.md

@@ -9,16 +9,12 @@ description: Protocol for cleaning parallel-coder worktree lanes before retry. T
 Before re-dispatching a coder for a task that already has a lane (any prior dispatch status).
 
 ## Protocol
-1. **Ownership check — do this FIRST, before any deletion.** Confirm the lane is not owned by another ACTIVE session: read `.swarm/session/state.json` and verify no other session's `delegationChains` reference `<session>/<task>`. If another active session owns it, STOP — surface the conflict and do not delete.
-2. **Check for existing lane branch:** `git branch --list "swarm/lane/<session>/<task>"`
-3. **If branch exists:**
-   - Verify 0-commits-ahead: `git log --oneline HEAD..swarm/lane/<session>/<task>` — empty = safe
-   - Delete: `git branch -d swarm/lane/<session>/<task>` (use `-D` only if `-d` fails AND the commits are confirmed unneeded)
-   - Under full-auto, `-D` is deny-pattern-blocked — use `-d`
-4. **Remove the per-lane worktree directory.** Target the SPECIFIC lane `.swarm-worktrees/<session>/<task>`, NOT the session parent (the parent holds sibling lanes). Prefer `git worktree remove .swarm-worktrees/<session>/<task>` — this is permitted because the path is under the `.swarm-worktrees/` base. A bare `rm -rf` / `Remove-Item -Recurse -Force` on `.swarm-worktrees/` is deny-pattern-blocked by the guardrail, so do not use it here.
-5. **Prune:** `git worktree prune`
-6. **Verify:** `git branch --list "swarm/lane/<session>/<task>"` returns empty
-7. Only after cleanup, proceed to `declare_scope` + coder dispatch
+1. **Prefer built-in provisioning cleanup.** Re-dispatch normally through the standard coder/worktree path. Provisioning pre-cleans stale same-lane worktrees/branches when ownership is safe and the existing lane is clean.
+2. **If provisioning blocks:** Treat the error as signal. Dirty lanes, lanes active in another worktree, and lanes owned by another active session must be surfaced to the user instead of deleted.
+3. **If manual cleanup is explicitly required:** Do the ownership check FIRST. Confirm the lane is not owned by another ACTIVE session: read `.swarm/session/state.json` and verify no other session's `delegationChains` reference `<session>/<task>`. If another active session owns it, STOP.
+4. **Remove only the specific lane.** Target `.swarm-worktrees/<session>/<task>`, never the session parent. Prefer `git worktree remove .swarm-worktrees/<session>/<task>` and then `git worktree prune`.
+5. **Delete only confirmed stale branches.** `git branch -d swarm/lane/<session>/<task>` is allowed after confirming the branch is not checked out and contains no needed commits. Use force deletion only with explicit human approval.
+6. **Verify:** `git branch --list "swarm/lane/<session>/<task>"` returns empty before retrying.
 
 ## Root cause
-The provisioning code should auto-clean after coder completion/denial/cancellation (tracked in issue #1746 item 1). This playbook is the temporary protocol.
+Stale same-lane worktrees and branches used to require manual cleanup before retry. Provisioning now handles the safe clean/stale cases automatically and fails closed for dirty, active, or cross-session-owned lanes.

package/dist/cli/{core-gjac7vd9.js → core-9a7nn51x.js}

@@ -14,7 +14,7 @@ import {
   resolveWorktreeBaseDir,
   shortenWorktreePath,
   writeLaneProfileToDiskReal
-} from "./index-dm35e55e.js";
+} from "./index-dnbca696.js";
 import"./index-5e4e2hvv.js";
 import"./index-p0arc26j.js";
 import"./index-zgwm4ryv.js";

package/dist/cli/{curator-qpgdv0fp.js → curator-2hm297yd.js}

@@ -12,7 +12,7 @@ import {
   runCuratorInit,
   runCuratorPhase,
   writeCuratorSummary
-} from "./index-smk3st51.js";
+} from "./index-7xfh4k71.js";
 import"./index-53z1afgh.js";
 import"./index-c8s9a3zh.js";
 import"./index-xddysq89.js";
@@ -32,7 +32,7 @@ import"./index-jt58s751.js";
 import"./index-wvrj16f0.js";
 import"./index-q1exe2b3.js";
 import"./index-vzzt3ygr.js";
-import"./index-dm35e55e.js";
+import"./index-dnbca696.js";
 import"./index-r3j3458j.js";
 import"./index-jtqkh8jf.js";
 import"./index-5e4e2hvv.js";

package/dist/cli/{curator-llm-factory-3x89ddtd.js → curator-llm-factory-19s0r5v6.js}

@@ -1,7 +1,7 @@
 // @bun
 import {
   createCuratorLLMDelegate
-} from "./index-smk3st51.js";
+} from "./index-7xfh4k71.js";
 import"./index-53z1afgh.js";
 import"./index-c8s9a3zh.js";
 import"./index-xddysq89.js";
@@ -21,7 +21,7 @@ import"./index-jt58s751.js";
 import"./index-wvrj16f0.js";
 import"./index-q1exe2b3.js";
 import"./index-vzzt3ygr.js";
-import"./index-dm35e55e.js";
+import"./index-dnbca696.js";
 import"./index-r3j3458j.js";
 import"./index-jtqkh8jf.js";
 import"./index-5e4e2hvv.js";

package/dist/cli/{guardrail-explain-jdwp6h9c.js → guardrail-explain-8w9rkdjx.js}

@@ -1,8 +1,8 @@
 // @bun
 import {
   handleGuardrailExplain
-} from "./index-4k3gh2nz.js";
-import"./index-smk3st51.js";
+} from "./index-9ptp9kky.js";
+import"./index-7xfh4k71.js";
 import"./index-53z1afgh.js";
 import"./index-c8s9a3zh.js";
 import"./index-xddysq89.js";
@@ -22,7 +22,7 @@ import"./index-jt58s751.js";
 import"./index-wvrj16f0.js";
 import"./index-q1exe2b3.js";
 import"./index-vzzt3ygr.js";
-import"./index-dm35e55e.js";
+import"./index-dnbca696.js";
 import"./index-r3j3458j.js";
 import"./index-jtqkh8jf.js";
 import"./index-5e4e2hvv.js";

package/dist/cli/{hive-promoter-rs5dr4pf.js → hive-promoter-0f8xh2rn.js}

@@ -5,7 +5,7 @@ import {
   isHiveEligible,
   promoteFromSwarm,
   promoteToHive
-} from "./index-smk3st51.js";
+} from "./index-7xfh4k71.js";
 import"./index-53z1afgh.js";
 import"./index-c8s9a3zh.js";
 import"./index-xddysq89.js";
@@ -25,7 +25,7 @@ import"./index-jt58s751.js";
 import"./index-wvrj16f0.js";
 import"./index-q1exe2b3.js";
 import"./index-vzzt3ygr.js";
-import"./index-dm35e55e.js";
+import"./index-dnbca696.js";
 import"./index-r3j3458j.js";
 import"./index-jtqkh8jf.js";
 import"./index-5e4e2hvv.js";