opencode-swarm 7.96.0 → 7.97.0

package/.opencode/skills/codebase-review-swarm/references/review-protocol-v8.2.md

@@ -128,7 +128,7 @@ Before writing under `.swarm/`, verify `.swarm/` is ignored or locally excluded.
 1. Run 0A alone.
 2. After 0A, run 0B and 0C through `dispatch_lanes_async` only if the repository is large enough to benefit. While those lanes run, the Architect continues deterministic inventory work that does not depend on their results.
 3. After 0B, run 0D and 0E through `dispatch_lanes_async` only if 0E can leave `linked_claims` blank for Architect linking in 0J. Otherwise run 0D before 0E.
-4. Preferred async batch order: batch 1 = 0F and 0G; batch 2 = 0H and 0I. Never exceed two Phase 0 agents.
+4. Preferred async batch order: batch 1 = 0F and 0G; batch 2 = 0H and 0I. Never exceed two Phase 0 agents — Phase 0 inventory units (0A→0J) form a largely sequential dependency chain, so concurrency is intentionally capped at 2 to respect that ordering rather than scaled toward the 8-lane dispatch limit.
 5. Run 0F after 0E when possible.
 6. Run 0G after 0B and 0C.
 7. Run 0H and 0I after 0B and 0C.

package/.opencode/skills/deep-dive/SKILL.md

@@ -18,7 +18,7 @@ Read-only deep audit of a specified codebase scope using parallel explorer waves
 Parse the MODE: DEEP_DIVE header to extract:
 - `scope`: the codebase area to audit (e.g., "auth", "payment flow", "src/hooks/")
 - `profile`: one of standard | security | ux | architecture | full (default: standard)
-- `max_explorers`: integer 1..8 (default: 6, or 8 for full profile)
+- `max_explorers`: integer 1..8 — upper bound on explorer waves (default: 6, or 8 for full profile). This is a CAP, not a fixed count: scale the actual wave size to the resolved scope surface — a trivial scope needs 1–2 explorers, a typical scope 3–5, a large multi-module scope up to the cap — never fix the count in advance.
 - `output`: markdown | json (default: markdown)
 - `update_main`: boolean (default: true) — whether to fetch/ff-only main before starting
 - `allow_dirty`: boolean (default: false) — whether to proceed with uncommitted changes
@@ -52,6 +52,8 @@ Dispatch explorer waves with `dispatch_lanes_async` when available. Each wave co
 - ~3500 total lines across all files in a mission
 - Group files by import proximity (files that import each other go in the same mission)
 
+**Partition is the contract:** missions own non-overlapping file sets — no file appears in two missions — and the union of all missions must cover every file in the Step 2 scope map. Any scope-map file not assigned to a mission is an explicit coverage gap, not an optional skip.
+
 **Profile-based lane selection — each profile activates specific lanes:**
 
 | Lane | Template | standard | security | ux | architecture | full |

package/.opencode/skills/deep-research/SKILL.md

@@ -123,14 +123,14 @@ synthesis lanes run, poll with `collect_lane_results` without `wait` (or
 `wait: false`) to process completed worker responses as they settle while
 continuing independent architect work between polls. Before Step 5, call
 `collect_lane_results` with `wait: true` for every open synthesis batch only if
-lanes are still pending and no independent work remains. Collect all completed
-worker responses into a candidate findings set, each finding tagged with its
-subtopic, evidence refs, and the worker's confidence. Treat missing, stale,
-cancelled, or failed lanes as explicit coverage gaps. If `dispatch_lanes_async`
-is unavailable, use blocking `dispatch_lanes` and record that async advisory
-lanes were unavailable; do not substitute per-agent Task calls for this fallback
-unless lane tools are unavailable and you explicitly verify equivalent agent
-type, prompt, scope, and isolation.
+lanes are still pending and no independent work remains. Do not advance to Step 5
+until every synthesis lane is settled. Collect all completed worker responses into
+a candidate findings set, each finding tagged with its subtopic, evidence refs,
+and the worker's confidence. Treat missing, stale, cancelled, or failed lanes as
+explicit coverage gaps. If `dispatch_lanes_async` is unavailable, use blocking
+`dispatch_lanes` and record that async advisory lanes were unavailable; do not
+substitute per-agent Task calls for this fallback unless lane tools are unavailable
+and you explicitly verify equivalent agent type, prompt, scope, and isolation.
 
 ## Step 5 — Dual-Reviewer Claim Verification
 

package/.opencode/skills/pre-phase-briefing/SKILL.md

@@ -42,15 +42,31 @@ This briefing is a HARD REQUIREMENT for ALL phases. Skipping it is a process vio
 
 ### CODEBASE REALITY CHECK (Required Before Speccing or Planning)
 
-Before any spec generation, plan creation, or plan ingestion begins, the Architect must dispatch the Explorer agent in targeted, scoped chunks — one per logical area of the codebase referenced by the work (e.g., per module, per hook, per config surface). Each chunk must be explored with full depth rather than a broad surface pass.
+Before any spec generation, plan creation, or plan ingestion begins, the Architect must verify the codebase reality of every item the work references. This runs as **asynchronous, fanned-out Explorer lanes by default**, joined behind a hard settlement gate — never as a single blocking explorer call, and never as fire-and-forget.
 
-For each scoped chunk, Explorer must determine:
+**1. Enumerate and partition the references (before dispatch).**
+List every referenced item — file, module, function, API, config surface, and behavioral assumption — named or implied by the spec, the user request, or the plan. Partition them into **non-overlapping** lane assignments. The partition is the contract: no two lanes may share a reference (this prevents duplicated work), and the **union of all lanes must cover every referenced item** (this prevents gaps). Under-specified lane boundaries are the dominant fan-out failure mode — be explicit about what each lane owns.
+
+**2. Scale the number of lanes to the size of the referenced surface.**
+- Trivial surface (a single file/function, one logical area) → **1 lane**.
+- Typical phase spanning a few areas → **2–4 lanes**.
+- Large surface (many modules/hooks/config surfaces) → **more lanes, up to the dispatch cap of 8 lanes per batch**.
+
+Do not fix the lane count in advance and do not over-spawn: extra lanes on a small surface waste tokens without improving coverage, while too few on a large surface leave gaps. Split by codebase area by default; when the surface is a single dense area, split by check-type instead — one lane for *existence & current state*, one for *assumption correctness & prior-work*.
+
+**3. Dispatch asynchronously, then keep working.**
+Dispatch the lanes with `dispatch_lanes_async`, record the returned `batch_id`, and continue **non-dependent** Architect work while they run — digest the retrospective and `user_directives`, review the spec/plan text for internal consistency, check governance/QA-gate config and the obligation ledger, and prepare the plan skeleton / task decomposition. This is dispatch-and-keep-busy, not fire-and-forget. Poll with `collect_lane_results` (wait omitted or false) to process settled lanes incrementally, or join with `wait: true` once independent work is exhausted.
+
+Each lane must be given: its objective, its named (disjoint) reference subset, the fixed REALITY-CHECK output format below, and clear boundaries. Lanes are read-only — they cannot `declare_scope` or mutate the worktree.
+
+**4. For each referenced item, the lane must determine:**
 - Does this file/module/function already exist?
 - If it exists, what is its current state? Does it already implement any part of what the plan or spec describes?
 - Is the plan's or user's assumption about the current state accurate? Flag any discrepancy between what is expected and what actually exists.
 - Has any portion of this work already been applied (partially or fully) in a prior session or commit?
 
-Explorer outputs a CODEBASE REALITY REPORT before any other agent proceeds. The report must list every referenced item with one of:
+**5. Hard settlement gate (join before any downstream work).**
+The Architect synthesizes the lane outputs into a single CODEBASE REALITY REPORT. The report must list every referenced item with one of:
   NOT STARTED | PARTIALLY DONE | ALREADY COMPLETE | ASSUMPTION INCORRECT
 
 Format:
@@ -59,11 +75,11 @@ Format:
     ✗ src/services/status-service.ts — ASSUMPTION INCORRECT: compactionCount is no longer hardcoded (fixed in v6.29.1)
     ✓ src/config/evidence-schema.ts — confirmed phase_number min(1)
 
-No implementation agent (coder, reviewer, test-engineer) may begin until this report is finalized.
+No spec finalization, plan generation, plan ingestion, `declare_scope`, or implementation-agent dispatch (coder, reviewer, test-engineer) may begin until ALL lanes in the batch are settled (`collect_lane_results` reports `all_settled`) AND this report is finalized. A lane that is missing, failed, or timed out is an explicit coverage gap, not a pass: mark the affected references BLOCKED or SKIPPED_WITH_REASON and resolve them before proceeding — never silently continue. Async dispatch changes *when* the Architect waits, never *whether* the gate holds.
 
 This check fires automatically in:
 - MODE: SPECIFY — before explorer dispatch for context (step 2)
 - MODE: PLAN — before plan generation or validation
 - EXTERNAL PLAN IMPORT PATH — before parsing the provided plan
 
-GREENFIELD EXEMPTION: If the work is purely greenfield (new project, no existing codebase references), skip this check.
+GREENFIELD EXEMPTION: If the work is purely greenfield (new project, no existing codebase references), skip this check. A trivial single-area surface stays a single lane rather than being force-fanned.

package/.opencode/skills/swarm-pr-feedback/SKILL.md

@@ -198,15 +198,27 @@ If a source is unavailable, retry with alternative access paths. If unavailable
 After the complete feedback ledger exists and before editing, use
 `dispatch_lanes_async` when available for independent read-only verification lanes:
 comment classification, CI/log root-cause inspection, test impact mapping,
-release/docs claim checks, and stale-branch/conflict analysis. Record each
-returned `batch_id`, then continue only ledger-safe architect work: normalize
-feedback IDs, gather deterministic PR metadata, prepare reproduction commands,
-and plan likely fix groups. Do not edit, close items, or mark feedback resolved
-from running lanes.
+release/docs claim checks, and stale-branch/conflict analysis. Partition the
+ledger so each `FB-###` item is owned by exactly one verification lane and the
+union of lanes covers the entire ledger — no feedback item may be left
+unassigned to a lane; state each lane's owned `FB-###` range in its prompt. Scale
+the lane count to the ledger size: a 1–3 item round may use a single combined
+lane, while a large multi-round intake may warrant one lane per category above.
+Cap each `dispatch_lanes_async` batch at 8 lanes (`MAX_LANES`); if the ledger
+needs more than 8 verification lanes, dispatch in sequential batches and settle
+each batch's COVERAGE GATE before the next — do not over-spawn lanes for a
+trivial round. Record each returned `batch_id`, then continue only ledger-safe
+architect work: normalize feedback IDs, gather deterministic PR metadata, prepare
+reproduction commands, and plan likely fix groups. Do not edit, close items, or
+mark feedback resolved from running lanes.
 
 Before the Verification step can mark any item `RESOLVED`, `DISPROVED`,
-`PRE_EXISTING`, `NEEDS_MORE_EVIDENCE`, or `NEEDS_USER_DECISION`, call
-`collect_lane_results` with `wait: true` for every open verification batch.
+`PRE_EXISTING`, `NEEDS_MORE_EVIDENCE`, or `NEEDS_USER_DECISION`, every open
+verification batch must be fully settled. Poll with `collect_lane_results` (wait
+omitted or `false`) to process settled lanes incrementally — clustering confirmed
+items and pre-reading files for settled findings while ledger-safe work remains —
+then issue a final `collect_lane_results` with `wait: true` per batch once
+independent work is exhausted, to confirm every lane is settled.
 Missing, stale, cancelled, or failed lanes are coverage gaps that must be closed
 before marking any item RESOLVED/DISPROVED/PRE_EXISTING. Apply the COVERAGE GATE:
 retry failed lanes (max 2), deploy a verified equivalent alternative (same agent

package/.opencode/skills/swarm-pr-review/SKILL.md

@@ -558,6 +558,8 @@ in the same batch unless intentionally replacing that exact lane before dispatch
 
 Explorers optimize for recall. Over-reporting is expected. Explorers produce candidates only.
 
+The six lanes are a fixed **check-type** partition (correctness / security / deps / docs / tests / performance), not an area partition: the count is intentionally constant — every PR needs all six review dimensions — and the lanes deliberately overlap by file, each receiving the same diff via `common_prompt` and viewing it through a different lens. This is the deliberate exception to surface-scaled fan-out: the base wave is a fixed six by design, never collapsed or expanded with the size of the change. Coverage is guaranteed by the six dimensions each reading the whole diff, not by partitioning files across lanes — so the disjoint-partition rule that governs area-split fan-outs does not apply to these check-type lanes.
+
 | Lane | Focus | Required checks |
 |---|---|---|
 | Lane 1: Correctness and edge cases | Logic errors, null/undefined handling, incorrect operators, async ordering, races, off-by-one, error paths | input domain, nullability, async/await, loop termination, exception behavior, backward compatibility |

package/dist/cli/{guardrail-explain-0hw3kyab.js → guardrail-explain-t6svvtmn.js}

@@ -1,8 +1,8 @@
 // @bun
 import {
   handleGuardrailExplain
-} from "./index-d3ds25vx.js";
-import"./index-q0t7gpf2.js";
+} from "./index-a9ghr5cx.js";
+import"./index-sgdr2e4n.js";
 import"./index-b223mczb.js";
 import"./index-2a6ppa65.js";
 import"./index-fjxjb66n.js";

package/dist/cli/{index-d3ds25vx.js → index-a9ghr5cx.js}

@@ -12,7 +12,7 @@ import {
   detectPosixWrites,
   detectWindowsWrites,
   resolveWriteTargets
-} from "./index-q0t7gpf2.js";
+} from "./index-sgdr2e4n.js";
 import {
   checkFileAuthority,
   classifyFile,

package/dist/cli/{index-r0zbs7ny.js → index-byb9tgay.js}

@@ -1,7 +1,7 @@
 // @bun
 import {
   handleGuardrailExplain
-} from "./index-d3ds25vx.js";
+} from "./index-a9ghr5cx.js";
 import {
   handleGuardrailLog
 } from "./index-rdc6nvmw.js";
@@ -78,7 +78,7 @@ import {
   handleWriteRetroCommand,
   normalizeSwarmCommandInput,
   resolveCommand
-} from "./index-q0t7gpf2.js";
+} from "./index-sgdr2e4n.js";
 import"./index-b223mczb.js";
 import"./index-2a6ppa65.js";
 import"./index-fjxjb66n.js";

package/dist/cli/{index-q0t7gpf2.js → index-sgdr2e4n.js}

@@ -909,7 +909,7 @@ var init_executor = __esm(() => {
 // package.json
 var package_default = {
   name: "opencode-swarm",
-  version: "7.96.0",
+  version: "7.97.0",
   description: "Architect-centric agentic swarm plugin for OpenCode - hub-and-spoke orchestration with SME consultation, code generation, and QA review",
   main: "dist/index.js",
   types: "dist/index.d.ts",
@@ -31857,7 +31857,7 @@ function buildDetailedHelp(commandName, entry) {
 async function handleHelpCommand(ctx) {
   const targetCommand = ctx.args.join(" ");
   if (!targetCommand) {
-    const { buildHelpText } = await import("./index-r0zbs7ny.js");
+    const { buildHelpText } = await import("./index-byb9tgay.js");
     return buildHelpText();
   }
   const tokens = targetCommand.split(/\s+/);
@@ -31866,7 +31866,7 @@ async function handleHelpCommand(ctx) {
     return _internals45.buildDetailedHelp(resolved.key, resolved.entry);
   }
   const similar = _internals45.findSimilarCommands(targetCommand);
-  const { buildHelpText: fullHelp } = await import("./index-r0zbs7ny.js");
+  const { buildHelpText: fullHelp } = await import("./index-byb9tgay.js");
   if (similar.length > 0) {
     return `Command '/swarm ${targetCommand}' not found.
 
@@ -31999,7 +31999,7 @@ var COMMAND_REGISTRY = {
   },
   "guardrail explain": {
     handler: async (ctx) => {
-      const { handleGuardrailExplain } = await import("./guardrail-explain-0hw3kyab.js");
+      const { handleGuardrailExplain } = await import("./guardrail-explain-t6svvtmn.js");
       return handleGuardrailExplain(ctx.directory, ctx.args);
     },
     description: "Dry-run: show what the guardrails would do to a command or write target (executes nothing)",

package/dist/cli/index.js

@@ -7,7 +7,7 @@ import {
   getPluginLockFilePaths,
   package_default,
   resolveCommand
-} from "./index-q0t7gpf2.js";
+} from "./index-sgdr2e4n.js";
 import"./index-b223mczb.js";
 import"./index-2a6ppa65.js";
 import"./index-fjxjb66n.js";

package/dist/index.js

@@ -69,7 +69,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "opencode-swarm",
-    version: "7.96.0",
+    version: "7.97.0",
     description: "Architect-centric agentic swarm plugin for OpenCode - hub-and-spoke orchestration with SME consultation, code generation, and QA review",
     main: "dist/index.js",
     types: "dist/index.d.ts",
@@ -102537,7 +102537,7 @@ Purpose: Read the previous retrospective and produce a codebase reality report b
 ACTION: Load skill file:.opencode/skills/pre-phase-briefing/SKILL.md immediately. Follow the protocol defined there.
 
 HARD CONSTRAINTS:
-- Complete the codebase reality report before starting or resuming phase implementation.
+- Complete the codebase reality report before spec finalization, plan generation, plan ingestion, declare_scope, or starting/resuming phase implementation. Dispatching the reality-check lanes asynchronously is allowed and preferred; settling all lanes before any of that downstream work is not optional.
 
 ### MODE: COUNCIL
 Activates when the user invokes /swarm council or requests a council-style decision review.

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "opencode-swarm",
-	"version": "7.96.0",
+	"version": "7.97.0",
 	"description": "Architect-centric agentic swarm plugin for OpenCode - hub-and-spoke orchestration with SME consultation, code generation, and QA review",
 	"main": "dist/index.js",
 	"types": "dist/index.d.ts",