opencode-swarm-plugin 0.43.0 → 0.44.1

package/CHANGELOG.md

@@ -1,2255 +0,0 @@
-# opencode-swarm-plugin
-
-## 0.43.0
-
-### Minor Changes
-
-- [`d74f68b`](https://github.com/joelhooks/swarm-tools/commit/d74f68ba491fdd127173c1993400c16b17479c3a) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔭 Observability Stack: See What Your Swarm Is Doing
-
-  ```
-      ╭──────────────────────────────────────────────────────────╮
-      │                                                          │
-      │   "Observability is about instrumenting your system      │
-      │    in a way that ensures sufficient information about    │
-      │    a system's runtime is collected and analyzed so       │
-      │    that when something goes wrong, it can help you       │
-      │    understand why."                                      │
-      │                                                          │
-      │                    — AI Engineering, Chip Huyen          │
-      │                                                          │
-      ╰──────────────────────────────────────────────────────────╯
-  ```
-
-  Five new modules for understanding multi-agent coordination at runtime:
-
-  ### Error Enrichment (`error-enrichment.ts`)
-
-  ```typescript
-  throw new SwarmError("File reservation failed", {
-    file: "src/auth.ts",
-    agent: "DarkHawk",
-    epic_id: "mjmas3zxlmg",
-    recent_events: [
-      /* last 5 events */
-    ],
-  });
-  ```
-
-  - `SwarmError` class with structured context (file, line, agent, epic, events)
-  - `enrichError()` wraps any error with swarm context
-  - `debugLog()` respects `DEBUG=swarm:*` patterns
-  - `suggestFix()` maps 8+ error patterns to actionable fixes
-
-  ### SQL Analytics (`swarm query`)
-
-  ```bash
-  swarm query --preset failed_decompositions
-  swarm query --sql "SELECT * FROM events WHERE type='worker_spawned'"
-  swarm query --preset duration_by_strategy --format csv
-  ```
-
-  10 preset queries: `failed_decompositions`, `duration_by_strategy`, `file_conflicts`, `worker_success_rate`, `review_rejections`, `blocked_tasks`, `agent_activity`, `event_frequency`, `error_patterns`, `compaction_stats`
-
-  ### Dashboard Data (`swarm dashboard`)
-
-  ```bash
-  swarm dashboard --epic mjmas3zxlmg --refresh 1000
-  ```
-
-  Real-time data fetching: worker status, subtask progress, file locks, recent messages, epic list.
-
-  ### Event Replay (`swarm replay`)
-
-  ```bash
-  swarm replay mjmas3zxlmg --speed 2x --type worker_spawned
-  swarm replay mjmas3zxlmg --agent DarkHawk --since "2025-12-25T10:00:00"
-  ```
-
-  Replay epic events with timing control. Filter by type, agent, time range. Debug coordination failures by watching the sequence unfold.
-
-  ### Export Formats (`swarm export`)
-
-  ```bash
-  swarm export --format otlp --epic mjmas3zxlmg  # OpenTelemetry traces
-  swarm export --format csv --output events.csv   # RFC 4180 compliant
-  swarm export --format json | jq '.[] | select(.type=="error")'
-  ```
-
-  **Test Coverage:** 225 tests (150 unit + 75 CLI integration)
-
-  **TDD Enforced:** RED cells first, GREEN cells second. Every function tested before implementation.
-
-## 0.42.9
-
-### Patch Changes
-
-- [`823987d`](https://github.com/joelhooks/swarm-tools/commit/823987d7b7ef57bf636665008ebbcdffe333e828) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Worktree Support + Graceful Degradation
-
-  ```
-      ┌─────────────────────────────────────────────────────┐
-      │                                                     │
-      │   "It is impossible to reduce the probability       │
-      │    of a fault to zero; therefore it is usually      │
-      │    best to design fault-tolerance mechanisms        │
-      │    that prevent faults from causing failures."      │
-      │                                                     │
-      │                    — Kleppmann, DDIA                │
-      │                                                     │
-      └─────────────────────────────────────────────────────┘
-  ```
-
-  ### Git Worktree Support (#52)
-
-  All worktrees now share the main repository's database. No more isolated state per worktree.
-
-  ```
-  BEFORE:                          AFTER:
-  main-repo/.opencode/swarm.db     main-repo/.opencode/swarm.db
-  worktree-1/.opencode/swarm.db    worktree-1/ ──┐
-  worktree-2/.opencode/swarm.db    worktree-2/ ──┼─→ main-repo/.opencode/
-                                   worktree-3/ ──┘
-  ```
-
-  **How it works:**
-
-  - Detects worktrees by checking if `.git` is a file (not directory)
-  - Parses `gitdir:` path to resolve main repo location
-  - All DB operations automatically use main repo's `.opencode/`
-
-  ### Graceful Degradation for semantic-memory (#53)
-
-  When Ollama is unavailable, `semantic-memory_find` now falls back to full-text search instead of erroring.
-
-  **Before:** `OllamaError: Connection failed` → tool fails
-  **After:** Warning logged → FTS results returned → tool works
-
-  Also added `repairStaleEmbeddings()` to fix the "dimensions are different: 0 != 1024" error when memories were stored without embeddings.
-
-  ### New Skill: gh-issue-triage
-
-  Added `.opencode/skills/gh-issue-triage/` for GitHub issue triage workflow:
-
-  - Extracts contributor profiles including Twitter handles
-  - Templates for acknowledgment comments
-  - Changeset credit templates with @mentions
-
-  ***
-
-  Thanks to [@justBCheung](https://x.com/justBCheung) for filing both issues with excellent debugging context. 🙏
-
-- Updated dependencies [[`823987d`](https://github.com/joelhooks/swarm-tools/commit/823987d7b7ef57bf636665008ebbcdffe333e828)]:
-  - swarm-mail@1.6.0
-
-## 0.42.8
-
-### Patch Changes
-
-- [`a797bea`](https://github.com/joelhooks/swarm-tools/commit/a797bea871e5d698ebb38b41f47ff07faa7c108b) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔗 Tweets Now Link to the Right PR
-
-  Release tweets were linking to the wrong PR. The old logic grabbed "most recent merged PR that isn't a version bump" - but with the new `release:` prefix on version PRs, it was picking up stale PRs.
-
-  **Fixed:** Now uses `github.sha` to find the exact PR that triggered the workflow. No more guessing.
-
-  ```
-  BEFORE: gh pr list --limit 5 --jq 'filter...'  → wrong PR
-  AFTER:  gh pr list --search "${{ github.sha }}" → triggering PR
-  ```
-
-## 0.42.7
-
-### Patch Changes
-
-- [`7a6a4a3`](https://github.com/joelhooks/swarm-tools/commit/7a6a4a37c4ea753de359dac5062d11186ee98ccd) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 📐 Swarm Insights Gets Its Blueprint
-
-  > _"The major documentation tool for information architecture... diagrams."_
-  > — Jesse James Garrett, The Elements of User Experience
-
-  The README now shows you how the swarm learns, not just that it does.
-
-  **Added:**
-
-  - ASCII diagram of the swarm learning loop (task → decompose → execute → complete → insights → repeat)
-  - Data flow architecture showing Event Store → Insights Aggregation → Agents
-  - Full API reference with TypeScript examples for coordinators and workers
-  - Token budget table (500 for coordinators, 300 for workers)
-  - Recommendation threshold table (≥80% = good, <40% = AVOID)
-  - Data sources table (Event Store, Semantic Memory, Anti-Pattern Registry)
-
-  **Why it matters:**
-  Diagrams > prose for architecture. Now you can see the feedback loop at a glance instead of reading paragraphs. The API examples are copy-pasteable.
-
-  ```
-  ┌──────────┐    ┌──────────┐    ┌──────────┐
-  │  TASK    │───▶│ INSIGHTS │───▶│  BETTER  │
-  │          │    │  LAYER   │    │  SWARMS  │
-  └──────────┘    └──────────┘    └──────────┘
-  ```
-
-- [`0ad79d5`](https://github.com/joelhooks/swarm-tools/commit/0ad79d57cd119517a8e04d0e74b4909f20a7f0be) Thanks [@joelhooks](https://github.com/joelhooks)! - ## Release Tweets Link to Source, PR Titles Get Smart
-
-  - Tweets now include link to the feature PR (or commit if pushed direct to main)
-  - Version bump PRs get AI-generated titles from changeset content via Opus
-  - No more "chore: update versions" - titles describe what actually shipped
-
-## 0.42.6
-
-### Patch Changes
-
-- [`ab90238`](https://github.com/joelhooks/swarm-tools/commit/ab902386883fa9654c9977d28888582fafc093e5) Thanks [@joelhooks](https://github.com/joelhooks)! - ## Query Epic Children Without Rawdogging JSONL
-
-  `hive_cells` and `hive_query` now support `parent_id` filter. Find all children of an epic in one call:
-
-  ```typescript
-  hive_cells({ parent_id: "epic-id" }); // Returns all subtasks
-  hive_query({ parent_id: "epic-id", status: "open" }); // Open subtasks only
-  ```
-
-  No more grep/jq on issues.jsonl. The tools do what they should.
-
-- Updated dependencies [[`ab90238`](https://github.com/joelhooks/swarm-tools/commit/ab902386883fa9654c9977d28888582fafc093e5)]:
-  - swarm-mail@1.5.5
-
-## 0.42.5
-
-### Patch Changes
-
-- [`45af762`](https://github.com/joelhooks/swarm-tools/commit/45af762ce656cf652847027d176d7bb7ff91f19b) Thanks [@joelhooks](https://github.com/joelhooks)! - ## The Bees Can Finally Tweet
-
-  New @swarmtoolsai app with proper OAuth. Releases now announce themselves.
-
-## 0.42.4
-
-### Patch Changes
-
-- [`4c7d869`](https://github.com/joelhooks/swarm-tools/commit/4c7d869e385677318fbbda7fa464bbe1223039f1) Thanks [@joelhooks](https://github.com/joelhooks)! - ## Switched to X API v2
-
-  Old action used deprecated v1.1 API. Now using direct OAuth 1.0a signed requests to v2 endpoint.
-
-## 0.42.3
-
-### Patch Changes
-
-- [`35eab3e`](https://github.com/joelhooks/swarm-tools/commit/35eab3e9e482b41e1535020a485561d5174e943c) Thanks [@joelhooks](https://github.com/joelhooks)! - ## First Real Tweet Incoming
-
-  Fixed X OAuth tokens - now with write permissions. Let's see if the bees can actually post.
-
-## 0.42.2
-
-### Patch Changes
-
-- [`9ded2a0`](https://github.com/joelhooks/swarm-tools/commit/9ded2a0929f430a3297e3b62858aa1143179542f) Thanks [@joelhooks](https://github.com/joelhooks)! - ## Tweet Bot Learns to Speak Swarm
-
-  Release tweets now use a manyshot prompt with examples that match the project's voice: terse, technical, slightly cheeky. Focus on what devs can DO, not what we shipped.
-
-## 0.42.1
-
-### Patch Changes
-
-- [`f6707d5`](https://github.com/joelhooks/swarm-tools/commit/f6707d53eb92021b6976212e903994c98c798483) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐦 @swarmtoolsai Now Tweets Releases
-
-  Automated release announcements are live! When packages publish to npm, Claude summarizes the changelog into a tweet and posts from @swarmtoolsai.
-
-  No more manual "hey we shipped" posts - the bees handle it now.
-
-## 0.42.0
-
-### Minor Changes
-
-- [`a79e04b`](https://github.com/joelhooks/swarm-tools/commit/a79e04b1bb3b40c09c5265b5d11739864799e4e2) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔭 Swarm Observability: See What Your Bees Are Doing
-
-  > "Observability is about instrumenting your system in a way that ensures sufficient information about a system's runtime is collected and analyzed so that when something goes wrong, it can help you understand why."
-  > — Chip Huyen, _AI Engineering_
-
-  New CLI commands to understand swarm health and history:
-
-  ### `swarm stats`
-
-  ```
-  ┌─────────────────────────────────────────┐
-  │        🐝  SWARM STATISTICS  🐝         │
-  ├─────────────────────────────────────────┤
-  │ Total Swarms: 42   Success: 87%         │
-  │ Avg Duration: 4.2min                    │
-  ├─────────────────────────────────────────┤
-  │ BY STRATEGY                             │
-  │ ├─ file-based      92% (23/25)          │
-  │ ├─ feature-based   78% (14/18)          │
-  │ ├─ risk-based      67% (2/3)            │
-  ├─────────────────────────────────────────┤
-  │ COORDINATOR HEALTH                      │
-  │ Violation Rate:   2%                    │
-  │ Spawn Efficiency: 94%                   │
-  │ Review Rate:      88%                   │
-  └─────────────────────────────────────────┘
-  ```
-
-  Options: `--since 24h/7d/30d`, `--json`
-
-  ### `swarm history`
-
-  Timeline of recent swarm activity with filtering:
-
-  - `--status success/failed/in_progress`
-  - `--strategy file-based/feature-based/risk-based`
-  - `--verbose` for subtask details
-
-  ### Prompt Insights Integration
-
-  Coordinators and workers now receive injected insights from past swarm outcomes:
-
-  - Strategy success rates as markdown tables
-  - Anti-pattern warnings for low-success strategies
-  - File/domain-specific learnings from semantic memory
-
-  This creates a feedback loop where swarms learn from their own history.
-
-  ### Also in this release
-
-  - **swarm-dashboard** (WIP): React/Vite visualizer scaffold
-  - **ADR-006**: Swarm PTY decision document
-  - **CI fix**: Smarter changeset detection prevents empty PR errors
-
-### Patch Changes
-
-- Updated dependencies [[`a79e04b`](https://github.com/joelhooks/swarm-tools/commit/a79e04b1bb3b40c09c5265b5d11739864799e4e2)]:
-  - swarm-mail@1.5.4
-
-## 0.41.0
-
-### Minor Changes
-
-- [`179b3f0`](https://github.com/joelhooks/swarm-tools/commit/179b3f0e49c7959f8d754c1274d301d0b3845a79) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Compaction Prompt Now Speaks Swarm
-
-  > _"Memory is essential for communication: we recall past interactions, infer preferences, and construct evolving mental models of those we engage with."_
-  > — Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory
-
-  When context compacts mid-swarm, coordinators were waking up confused. They had state information but no protocol guidance. Now the compaction prompt includes a condensed version of the swarm command template.
-
-  **What's New:**
-
-  The `SWARM_COMPACTION_CONTEXT` now includes:
-
-  1. **What Good Looks Like** - Behavioral examples showing ideal coordinator behavior
-
-     - ✅ Spawned researcher for unfamiliar tech → got summary → stored in semantic-memory
-     - ✅ Checked inbox every 5-10 minutes → caught blocked worker → unblocked in 2min
-     - ❌ Called context7 directly → dumped 50KB → context exhaustion
-
-  2. **Mandatory Behaviors Checklist** - Post-compaction protocol
-     - Inbox monitoring (every 5-10 min with intervention triggers)
-     - Skill loading (before spawning workers)
-     - Worker review (after every worker returns, 3-strike rule)
-     - Research spawning (never call context7/pdf-brain directly)
-
-  **Why This Matters:**
-
-  Coordinators resuming from compaction now have:
-
-  - Clear behavioral guidance (not just state)
-  - Actionable tool call examples
-  - Anti-patterns to avoid
-  - The same protocol as fresh `/swarm` invocations
-
-  **Backward Compatible:** Existing compaction hooks continue to work. This adds guidance, doesn't change the hook signature.
-
-### Patch Changes
-
-- [`3e7c126`](https://github.com/joelhooks/swarm-tools/commit/3e7c126b11aa6ad909ebcb2ab3cf77883f9acfe4) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🧪 Bulletproof Test Suite
-
-  > "Setting up our tests to run synchronously and using mocking libraries will greatly speed up our testing"
-  > — ng-book
-
-  Fixed test isolation issues that caused 19 tests to fail when run together but pass in isolation.
-
-  ### The Culprits
-
-  **1. Global fetch pollution** (`ollama.test.ts`)
-
-  ```typescript
-  // BEFORE: Replaced global.fetch, never restored it
-  global.fetch = mockFetch;
-
-  // AFTER: Save and restore
-  const originalFetch = global.fetch;
-  afterEach(() => {
-    global.fetch = originalFetch;
-  });
-  ```
-
-  **2. Port conflicts** (`durable-server.test.ts`)
-
-  - Tests used hardcoded ports (4483, 4484, 4485)
-  - Parallel test runs fought over the same ports
-  - Fixed: Use `port: 0` for OS-assigned ports, made `server.url` a getter
-
-  **3. AI SDK schema incompatibility** (`memory-operations.ts`)
-
-  - `z.discriminatedUnion` creates `oneOf` at top level
-  - Anthropic API requires `type: object` at top level
-  - Fixed: Flat object schema with optional fields
-
-  ### Test Stats
-
-  ```
-  Before: 19 failures when run together
-  After:  0 failures, 1406 tests pass
-  ```
-
-  ### Files Changed
-
-  - `src/memory/ollama.test.ts` - Restore global.fetch after each test
-  - `src/streams/durable-server.ts` - Dynamic port getter
-  - `src/streams/durable-server.test.ts` - Use port 0, rewrite for isolation
-  - `src/memory/memory-operations.ts` - Flat schema for Anthropic compatibility
-  - Renamed `memory-operations.test.ts` → `memory-operations.integration.test.ts`
-
-- Updated dependencies [[`3e7c126`](https://github.com/joelhooks/swarm-tools/commit/3e7c126b11aa6ad909ebcb2ab3cf77883f9acfe4)]:
-  - swarm-mail@1.5.3
-
-## 0.40.0
-
-### Minor Changes
-
-- [`948e031`](https://github.com/joelhooks/swarm-tools/commit/948e0318fe5e2c1a5d695a56533fc2a2a7753887) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔭 Observability Swarm: See What the Bees Are Doing
-
-  > "The unexamined swarm is not worth coordinating." — Socrates, probably
-
-  Four parallel workers descended on the observability stack and emerged victorious. The compaction hook no longer runs in darkness, coordinator sessions are now viewable, and the docs finally explain what all those JSONL files are for.
-
-  ### What's New
-
-  **Compaction Observability** (`src/compaction-observability.ts`)
-
-  - Metrics collector tracks phases: START → GATHER → DETECT → INJECT → COMPLETE
-  - Pattern extraction/skipping with reasons ("why didn't this get captured?")
-  - Timing breakdown per phase (analysis vs extraction vs storage)
-  - 15 tests (11 unit + 4 integration)
-
-  **`swarm log sessions` CLI**
-
-  - `swarm log sessions` — list all captured coordinator sessions
-  - `swarm log sessions <id>` — view events for a session (partial ID matching)
-  - `swarm log sessions --latest` — quick access to most recent
-  - `--type`, `--since`, `--limit`, `--json` filters
-  - 64 tests covering parsing, listing, filtering
-
-  **Coordinator Observability Docs**
-
-  - AGENTS.md: overview with quick commands
-  - evals/README.md: deep dive with ASCII flow diagrams, event type reference, JSONL examples, jq recipes
-
-  **Research: Coordinator Prompt Eval** (`.hive/analysis/coordinator-prompt-eval-research.md`)
-
-  - 26KB analysis of prompt iteration strategies
-  - Recommends: versioning + evalite (defer LLM-as-Judge to v0.34+)
-  - Implementation plan with effort estimates
-
-  ### The Observability Story
-
-  ```
-  CAPTURE ──────────► VIEW ──────────► SCORE
-  (eval-capture.ts)   (swarm log       (coordinator
-                       sessions)        evals)
-  ```
-
-  Now you can answer:
-
-  - "What did the last 10 compaction runs extract?"
-  - "Why didn't this pattern get captured?"
-  - "Which coordinator sessions had violations?"
-
-## 0.39.1
-
-### Patch Changes
-
-- [`19a6557`](https://github.com/joelhooks/swarm-tools/commit/19a6557cee9878858e7f61e2aba86b37a3ec10ad) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Eval Quality Gates: Signal Over Noise
-
-  The eval system now filters coordinator sessions to focus on high-quality data.
-
-  **Problem:** 67 of 82 captured sessions had <3 events - noise from aborted runs, test pokes, and incomplete swarms. This diluted eval scores and made metrics unreliable.
-
-  **Solution:** Quality filters applied BEFORE sampling:
-
-  | Filter               | Default | Purpose                           |
-  | -------------------- | ------- | --------------------------------- |
-  | `minEvents`          | 3       | Skip incomplete/aborted sessions  |
-  | `requireWorkerSpawn` | true    | Ensure coordinator delegated work |
-  | `requireReview`      | true    | Ensure full swarm lifecycle       |
-
-  **Impact:**
-
-  - Filters 93 noisy sessions automatically
-  - Overall eval score: 63% → 71% (true signal, not diluted)
-  - Coordinator discipline: 47% → 57% (accurate measurement)
-
-  **Usage:**
-
-  ```typescript
-  // Default: high-quality sessions only
-  const sessions = await loadCapturedSessions();
-
-  // Override for specific analysis
-  const allSessions = await loadCapturedSessions({
-    minEvents: 1,
-    requireWorkerSpawn: false,
-    requireReview: false,
-  });
-  ```
-
-  Includes 7 unit tests covering filter logic and edge cases.
-
-## 0.39.0
-
-### Minor Changes
-
-- [`aa12943`](https://github.com/joelhooks/swarm-tools/commit/aa12943f3edc8d5e23878b22f44073e4c71367c5) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Eval-Driven Development: The System That Scores Itself
-
-  > "What gets measured gets managed." — Peter Drucker
-  > "What gets scored gets improved." — The Swarm
-
-  The plugin now evaluates its own output quality through a progressive gate system. Every compaction prompt gets scored, tracked, and learned from. Regressions become impossible to ignore.
-
-  ### The Pipeline
-
-  ```
-  CAPTURE → SCORE → STORE → GATE → LEARN → IMPROVE
-     ↑                                      ↓
-     └──────────────────────────────────────┘
-  ```
-
-  ### What's New
-
-  **Event Capture** (5 integration points)
-
-  - `detection_triggered` - When compaction is detected
-  - `prompt_generated` - Full LLM prompt captured
-  - `context_injected` - Final content before injection
-  - All events stored to `~/.config/swarm-tools/sessions/{session_id}.jsonl`
-
-  **5 Compaction Prompt Scorers**
-
-  - `epicIdSpecificity` - Real IDs, not placeholders (20%)
-  - `actionability` - Specific tool calls with values (20%)
-  - `coordinatorIdentity` - ASCII header + mandates (25%)
-  - `forbiddenToolsPresent` - Lists what NOT to do (15%)
-  - `postCompactionDiscipline` - First tool is correct (20%)
-
-  **Progressive Gates**
-  | Phase | Threshold | Behavior |
-  |-------|-----------|----------|
-  | Bootstrap | N/A | Always pass, building baseline |
-  | Stabilization | 0.6 | Warn but pass |
-  | Production | 0.7 | Fail CI on regression |
-
-  **CLI Commands**
-
-  ```bash
-  swarm eval status          # Current phase, thresholds, scores
-  swarm eval history         # Trends with sparklines ▁▂▃▄▅▆▇█
-  swarm eval run [--ci]      # Execute evals, gate check
-  ```
-
-  **CI Integration**
-
-  - Runs after tests pass
-  - Posts results as PR comment with emoji status
-  - Only fails in production phase with actual regression
-
-  **Learning Feedback Loop**
-
-  - Significant score drops auto-stored to semantic memory
-  - Future agents learn from past failures
-  - Pattern maturity tracking
-
-  ### Breaking Changes
-
-  None. All new functionality is additive.
-
-  ### Files Changed
-
-  - `src/eval-capture.ts` - Event capture with Zod schemas
-  - `src/eval-gates.ts` - Progressive gate logic
-  - `src/eval-history.ts` - Score tracking over time
-  - `src/eval-learning.ts` - Failure-to-learning extraction
-  - `src/compaction-prompt-scoring.ts` - 5 pure scoring functions
-  - `evals/compaction-prompt.eval.ts` - Evalite integration
-  - `bin/swarm.ts` - CLI commands
-  - `.github/workflows/ci.yml` - CI integration
-
-  ### Test Coverage
-
-  - 422 new tests for eval-capture
-  - 48 CLI tests
-  - 7 integration tests for capture wiring
-  - All existing tests still passing
-
-### Patch Changes
-
-- Updated dependencies [[`aa12943`](https://github.com/joelhooks/swarm-tools/commit/aa12943f3edc8d5e23878b22f44073e4c71367c5)]:
-  - swarm-mail@1.5.2
-
-## 0.38.0
-
-### Minor Changes
-
-- [`41a1965`](https://github.com/joelhooks/swarm-tools/commit/41a19657b252eb1c7a7dc82bc59ab13589e8758f) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Coordinators Now Delegate Research to Workers
-
-  Coordinators finally know their place. They orchestrate, they don't fetch.
-
-  **The Problem:**
-  Coordinators were calling `repo-crawl_file`, `webfetch`, `context7_*` directly, burning expensive Sonnet context on raw file contents instead of spawning researcher workers.
-
-  **The Fix:**
-
-  ### Forbidden Tools Section
-
-  COORDINATOR_PROMPT now explicitly lists tools coordinators must NEVER call:
-
-  - `repo-crawl_*`, `repo-autopsy_*` - repository fetching
-  - `webfetch`, `fetch_fetch` - web fetching
-  - `context7_*` - library documentation
-  - `pdf-brain_search`, `pdf-brain_read` - knowledge base
-
-  ### Phase 1.5: Research Phase
-
-  New workflow phase between Initialize and Knowledge Gathering:
-
-  ```
-  swarm_spawn_researcher(
-    research_id="research-nextjs-cache",
-    tech_stack=["Next.js 16 Cache Components"],
-    project_path="/path/to/project"
-  )
-  ```
-
-  ### Strong Coordinator Identity Post-Compaction
-
-  When context compacts, the resuming agent now sees:
-
-  ```
-  ┌─────────────────────────────────────────────────────────────┐
-  │             🐝  YOU ARE THE COORDINATOR  🐝                 │
-  │             NOT A WORKER. NOT AN IMPLEMENTER.               │
-  │                  YOU ORCHESTRATE.                           │
-  └─────────────────────────────────────────────────────────────┘
-  ```
-
-  ### runResearchPhase Returns Spawn Instructions
-
-  ```typescript
-  const result = await runResearchPhase(task, projectPath);
-  // result.spawn_instructions = [
-  //   { research_id, tech, prompt, subagent_type: "swarm/researcher" }
-  // ]
-  ```
-
-  **32+ new tests, all 425 passing.**
-
-- [`b06f69b`](https://github.com/joelhooks/swarm-tools/commit/b06f69bc3db099c14f712585d88b42c801123d01) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔬 Eval Capture Pipeline: Complete
-
-  > "The purpose of computing is insight, not numbers." — Richard Hamming
-
-  Wire all eval-capture functions into the swarm execution path, enabling ground-truth collection from real swarm executions.
-
-  **What changed:**
-
-  | Function                  | Wired Into                     | Purpose                            |
-  | ------------------------- | ------------------------------ | ---------------------------------- |
-  | `captureDecomposition()`  | `swarm_validate_decomposition` | Records task → subtasks mapping    |
-  | `captureSubtaskOutcome()` | `swarm_complete`               | Records per-subtask execution data |
-  | `finalizeEvalRecord()`    | `swarm_record_outcome`         | Computes aggregate metrics         |
-
-  **New npm scripts:**
-
-  ```bash
-  bun run eval:run           # Run all evals
-  bun run eval:decomposition # Decomposition quality
-  bun run eval:coordinator   # Coordinator discipline
-  ```
-
-  **Data flow:**
-
-  ```
-  swarm_decompose → captureDecomposition → .opencode/eval-data.jsonl
-         ↓
-  swarm_complete → captureSubtaskOutcome → updates record with outcomes
-         ↓
-  swarm_record_outcome → finalizeEvalRecord → computes scope_accuracy, time_balance
-         ↓
-  evalite → reads JSONL → scores decomposition quality
-  ```
-
-  **Why it matters:**
-
-  - Enables data-driven decomposition strategy selection
-  - Tracks which strategies work for which task types
-  - Provides ground truth for Evalite evals
-  - Foundation for learning from swarm outcomes
-
-  **Key discovery:** New cell ID format doesn't follow `epicId.subtaskNum` pattern. Must use `cell.parent_id` to get epic ID for subtasks.
-
-### Patch Changes
-
-- [`56e5d4c`](https://github.com/joelhooks/swarm-tools/commit/56e5d4c5ac96ddd2184d12c63e163bb9c291fb69) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔬 Eval Capture Pipeline: Phase 1
-
-  > "The first step toward wisdom is getting things right. The second step is getting them wrong in interesting ways." — Marvin Minsky
-
-  Wire `captureDecomposition()` into `swarm_validate_decomposition` to record decomposition inputs/outputs for evaluation.
-
-  **What changed:**
-
-  - `swarm_validate_decomposition` now calls `captureDecomposition()` after successful validation
-  - Captures: epicId, projectPath, task, context, strategy, epicTitle, subtasks
-  - Data persisted to `.opencode/eval-data.jsonl` for Evalite consumption
-
-  **Why it matters:**
-
-  - Enables ground-truth collection from real swarm executions
-  - Foundation for decomposition quality evals
-  - Tracks what strategies work for which task types
-
-  **Tests added:**
-
-  - Verifies `captureDecomposition` called with correct params on success
-  - Verifies NOT called on validation failure
-  - Handles optional context/description fields
-
-  **Next:** Wire `captureSubtaskOutcome()` and `finalizeEvalRecord()` to complete the pipeline.
-
-## 0.37.0
-
-### Minor Changes
-
-- [`66b5795`](https://github.com/joelhooks/swarm-tools/commit/66b57951e2c114702c663b98829d5f7626607a16) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 `swarm cells` - Query Your Hive Like a Pro
-
-  New CLI command AND plugin tool for querying cells directly from the database.
-
-  ### CLI: `swarm cells`
-
-  ```bash
-  swarm cells                      # List all cells (table format)
-  swarm cells --status open        # Filter by status
-  swarm cells --type bug           # Filter by type
-  swarm cells --ready              # Next unblocked cell
-  swarm cells mjkmd                # Partial ID lookup
-  swarm cells --json               # Raw JSON for scripting
-  ```
-
-  **Replaces:** The awkward `swarm tool hive_query --json '{"status":"open"}'` pattern.
-
-  ### Plugin Tool: `hive_cells`
-
-  ```typescript
-  // Agents can now query cells directly
-  hive_cells({ status: "open", type: "task" });
-  hive_cells({ id: "mjkmd" }); // Partial ID works!
-  hive_cells({ ready: true }); // Next unblocked
-  ```
-
-  **Why this matters:**
-
-  - Reads from DATABASE (fast, indexed) not JSONL files
-  - Partial ID resolution built-in
-  - Consistent JSON array output
-  - Rich descriptions encourage agentic use
-
-  ### Also Fixed
-
-  - `swarm_review_feedback` tests updated for coordinator-driven retry architecture
-  - 425 tests passing
-
-## 0.36.1
-
-### Patch Changes
-
-- [`9c1f3f3`](https://github.com/joelhooks/swarm-tools/commit/9c1f3f3e7204f02c133c4a036fa34e83d8376a8c) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Coordinator Discipline: Prohibition-First Enforcement
-
-  Coordinators kept "just doing it themselves" after compaction. Now they can't.
-
-  **The Problem:**
-  After context compaction, coordinators would ignore their own instructions to "spawn workers for remaining subtasks" and edit files directly. The compaction context was narrative ("do this") rather than prescriptive ("NEVER do that").
-
-  **The Fix:**
-
-  ### 1. Prohibition-First Compaction Context
-
-  The `SWARM_COMPACTION_CONTEXT` now leads with explicit anti-patterns:
-
-  ```markdown
-  ### ⛔ NEVER DO THESE (Coordinator Anti-Patterns)
-
-  - ❌ **NEVER** use `edit` or `write` tools - SPAWN A WORKER
-  - ❌ **NEVER** run tests with `bash` - SPAWN A WORKER
-  - ❌ **NEVER** implement features yourself - SPAWN A WORKER
-  - ❌ **NEVER** "just do it myself to save time" - NO. SPAWN A WORKER.
-  ```
-
-  ### 2. Runtime Violation Detection
-
-  `detectCoordinatorViolation()` is now wired up in `tool.execute.before`:
-
-  - Detects when coordinators call `edit`, `write`, or test commands
-  - Emits warnings to help coordinators self-correct
-  - Captures VIOLATION events for post-hoc analysis
-
-  ### 3. Coordinator Context Tracking
-
-  New functions track when we're in coordinator mode:
-
-  - `setCoordinatorContext()` - Activated when `hive_create_epic` or `swarm_decompose` is called
-  - `isInCoordinatorContext()` - Checks if we're currently coordinating
-  - `clearCoordinatorContext()` - Cleared when epic is closed
-
-  **Why This Matters:**
-
-  Coordinators that do implementation work burn context, create conflicts, and defeat the purpose of swarm coordination. This fix makes the anti-pattern visible and provides guardrails to prevent it.
-
-  **Validation:**
-
-  - Check `~/.config/swarm-tools/sessions/` for VIOLATION events
-  - Run `coordinator-behavior.eval.ts` to score coordinator discipline
-
-- [`4c23c7a`](https://github.com/joelhooks/swarm-tools/commit/4c23c7a31013bc6537d83a9294b51540056cde93) Thanks [@joelhooks](https://github.com/joelhooks)! - ## Fix Double Hook Registration
-
-  The compaction hook was firing twice per compaction event because OpenCode's plugin loader
-  calls ALL exports as plugin functions. We were exporting `SwarmPlugin` as both:
-
-  1. Named export: `export const SwarmPlugin`
-  2. Default export: `export default SwarmPlugin`
-
-  This caused the plugin to register twice, doubling all hook invocations.
-
-  **Fix:** Changed to default-only export pattern:
-
-  - `src/index.ts`: `const SwarmPlugin` (no export keyword)
-  - `src/plugin.ts`: `export default SwarmPlugin` (no named re-export)
-
-  **Impact:** Compaction hooks now fire once. LLM calls during compaction reduced by 50%.
-
-- Updated dependencies [[`e0c422d`](https://github.com/joelhooks/swarm-tools/commit/e0c422de3f5e15c117cc0cc655c0b03242245be4), [`43c8c93`](https://github.com/joelhooks/swarm-tools/commit/43c8c93ef90b2f04ce59317192334f69d7c4204e)]:
-  - swarm-mail@1.5.1
-
-## 0.36.0
-
-### Minor Changes
-
-- [`ae213aa`](https://github.com/joelhooks/swarm-tools/commit/ae213aa49be977e425e0a767b5b2db16e462f76b) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔬 Compaction Hook: Now With X-Ray Vision
-
-  The compaction hook was logging to `console.log` like a caveman. Now it writes structured JSON logs to `~/.config/swarm-tools/logs/compaction.log` - visible via `swarm log compaction`.
-
-  **The Problem:**
-
-  - Plugin wrapper used `console.log` → stdout → invisible
-  - npm package had pino logging → but wrapper didn't use it
-  - Running `/compact` gave zero visibility into what happened
-
-  **The Fix:**
-  Added comprehensive file-based logging throughout the compaction flow:
-
-  ```
-  ┌─────────────────────────────────────────────────────────────┐
-  │                    COMPACTION LOGGING                       │
-  ├─────────────────────────────────────────────────────────────┤
-  │  compaction_hook_invoked     │ Full input/output objects    │
-  │  detect_swarm_*              │ CLI calls, cells, confidence │
-  │  query_swarm_state_*         │ Epic/subtask extraction      │
-  │  generate_compaction_prompt_*│ LLM timing, success/failure  │
-  │  context_injected_via_*      │ Which API used               │
-  │  compaction_complete_*       │ Final result + timing        │
-  └─────────────────────────────────────────────────────────────┘
-  ```
-
-  **Also Enhanced:**
-
-  - SDK message scanning for precise swarm state extraction
-  - Merged scanned state (ground truth) with hive detection (heuristic)
-  - 9 new tests for `scanSessionMessages()` (32 total passing)
-
-  **To See It Work:**
-
-  ```bash
-  swarm setup --reinstall  # Regenerate plugin wrapper
-  # Run /compact in OpenCode
-  swarm log compaction     # See what happened
-  ```
-
-### Patch Changes
-
-- [`5cfc42e`](https://github.com/joelhooks/swarm-tools/commit/5cfc42e93d3e5424e308857a40af4fd9fbda0ba3) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Swarm Workers Unchained
-
-  Removed the vestigial `max_subtasks` parameter from decomposition tools. It was dead code - the prompts already say "as many as needed" and the replacement was doing nothing.
-
-  **What changed:**
-
-  - Removed `max_subtasks` arg from `swarm_decompose`, `swarm_plan_prompt`, `swarm_delegate_planning`
-  - Removed from `DecomposeArgsSchema`
-  - Renamed `max_subtasks` → `subtask_count` in eval capture (records actual count, not a limit)
-  - Cleaned up tests that were passing the unused parameter
-
-  **Why it matters:**
-  The LLM decides how many subtasks based on task complexity, not an arbitrary cap. "Plan aggressively" means spawn as many workers as the task needs.
-
-  **No functional change** - the parameter wasn't being used anyway.
-
-## 0.35.0
-
-### Minor Changes
-
-- [`084f888`](https://github.com/joelhooks/swarm-tools/commit/084f888fcac4912f594428b1ac7148c8a8aaa422) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 👁️ Watch Your Swarm in Real-Time
-
-  `swarm log` now has a `--watch` mode for continuous log monitoring. No more running the command repeatedly - just sit back and watch the bees work.
-
-  ```bash
-  # Watch all logs
-  swarm log --watch
-
-  # Watch with filters
-  swarm log compaction -w --level error
-
-  # Faster polling (500ms instead of default 1s)
-  swarm log --watch --interval 500
-  ```
-
-  **New flags:**
-
-  - `--watch`, `-w` - Enable continuous monitoring mode
-  - `--interval <ms>` - Poll interval in milliseconds (default: 1000, min: 100)
-
-  **How it works:**
-
-  - Shows initial logs (last N lines based on `--limit`)
-  - Polls log files for new entries at the specified interval
-  - Tracks file positions for efficient incremental reads
-  - Handles log rotation gracefully (detects file truncation)
-  - All existing filters work: `--level`, `--since`, module name
-  - Clean shutdown on Ctrl+C
-
-  _"The hive that watches itself, debugs itself."_
-
-## 0.34.0
-
-### Minor Changes
-
-- [`704c366`](https://github.com/joelhooks/swarm-tools/commit/704c36690fb6fd52cfb9222ddeef3b663dfdb9ed) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🪵 Pino Logging Infrastructure
-
-  > "You can't improve what you can't measure." — Peter Drucker
-
-  Finally, visibility into what the swarm is actually doing.
-
-  ### What's New
-
-  **Structured Logging with Pino**
-
-  - Daily log rotation via `pino-roll` (14-day retention)
-  - Logs to `~/.config/swarm-tools/logs/`
-  - Module-specific log files (e.g., `compaction.1log`, `swarm.1log`)
-  - Pretty mode for development: `SWARM_LOG_PRETTY=1`
-
-  **Compaction Hook Instrumented**
-
-  - 14 strategic log points across all phases
-  - START: session context, trigger reason
-  - GATHER: per-source timing (hive, swarm-mail, skills)
-  - DETECT/INJECT: confidence scores, context decisions
-  - COMPLETE: duration, success, what was injected
-
-  **New CLI: `swarm log`**
-
-  ```bash
-  swarm log                    # Tail recent logs
-  swarm log compaction         # Filter by module
-  swarm log --level warn       # Filter by severity
-  swarm log --since 1h         # Last hour only
-  swarm log --json | jq        # Pipe to jq for analysis
-  ```
-
-  ### Why This Matters
-
-  The compaction hook does a LOT of work with zero visibility:
-
-  - Context injection decisions
-  - Data gathering from multiple sources
-  - Template rendering and size calculations
-
-  Now you can answer: "What did compaction do on the last run?"
-
-  ### Technical Details
-
-  - Pino + pino-roll for async, non-blocking file writes
-  - Child loggers for module namespacing
-  - Lazy initialization pattern for test isolation
-  - 56 new tests (10 logger + 18 compaction + 28 CLI)
-
-  Complements existing `DEBUG=swarm:*` env var approach — Pino for structured file logs, debug for stderr filtering.
-
-### Patch Changes
-
-- [`b5792bd`](https://github.com/joelhooks/swarm-tools/commit/b5792bd5f6aa4bf3ad9757fe351bc144e84f09af) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🎯 Coordinators Remember Who They Are
-
-  Fixed the compaction bug where coordinators lost their identity after context compression.
-
-  **The Problem:**
-  After compaction, coordinators would wake up and start doing worker tasks directly (running tests, editing files) instead of spawning workers. The injected context said "you are a coordinator" but gave worker-style resume commands.
-
-  **The Fix:**
-  `buildDynamicSwarmState()` now generates coordinator-focused context:
-
-  ```
-  ## 🎯 YOU ARE THE COORDINATOR
-
-  **Primary role:** Orchestrate workers, review their output, unblock dependencies.
-  **Spawn workers** for implementation tasks - don't do them yourself.
-
-  **RESUME STEPS:**
-  1. Check swarm status: `swarm_status(epic_id="bd-actual-id", ...)`
-  2. Check inbox: `swarmmail_inbox(limit=5)`
-  3. For in_progress subtasks: Review with `swarm_review`
-  4. For open subtasks: Spawn workers with `swarm_spawn_subtask`
-  5. For blocked subtasks: Investigate and unblock
-  ```
-
-  Also captures specific swarm state during detection:
-
-  - Epic ID and title (not placeholders)
-  - Subtask counts by status
-  - Actual project path
-
-  **New eval infrastructure:**
-
-  - `coordinator-behavior.eval.ts` - LLM-as-judge eval testing whether Claude actually behaves like a coordinator given the injected context
-  - Scorers for coordinator tools, avoiding worker behaviors, and coordinator mindset
-
-  > "The coordinator's job is to keep the swarm cooking, not to cook themselves."
-
-- Updated dependencies [[`a78a40d`](https://github.com/joelhooks/swarm-tools/commit/a78a40de32eb34d1738b208f2a36929a4ab6cb81), [`5a7c084`](https://github.com/joelhooks/swarm-tools/commit/5a7c084514297b5b9ca5df9459a74f18eb805b8a)]:
-  - swarm-mail@1.5.0
-
-## 0.33.0
-
-### Minor Changes
-
-- [`c41abcf`](https://github.com/joelhooks/swarm-tools/commit/c41abcfa37292b72fe41e0cf9d25c6612ae75fa2) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🎓 Skills Grow Up: Discovery Moves to OpenCode
-
-  > _"The best code is no code at all. Every new line of code you willingly bring into the world is code that has to be debugged, code that has to be read and understood, code that has to be supported."_
-  > — Jeff Atwood
-
-  Skills outgrew the nest. OpenCode is shipping native skills support following the [Agent Skills spec](https://spec.agentskills.com/), and our discovery tools are now redundant. Time to deprecate the scaffolding and let the platform handle what it does best.
-
-  ### What Changed
-
-  **Deprecated Tools** (soft deprecation with console warnings):
-
-  - `skills_list` - OpenCode will handle discovery natively
-  - `skills_use` - OpenCode will handle loading via `use skill <name>` syntax
-  - `skills_read` - OpenCode will handle resource access transparently
-  - `skills_execute` - OpenCode will handle script execution in skill context
-
-  **Authoring Tools Kept** (fully functional, no changes):
-
-  - `skills_create` - Create new skills with SKILL.md template
-  - `skills_update` - Update existing skill content
-  - `skills_init` - Initialize skills directory in projects
-  - `skills_add_script` - Add executable scripts to skills
-  - `skills_delete` - Remove project skills
-
-  **Bundled Skills** - All 6 global skills remain intact and spec-compliant:
-
-  - `testing-patterns` - Feathers seams + Beck's 4 rules
-  - `swarm-coordination` - Multi-agent task orchestration
-  - `cli-builder` - Command-line interface patterns
-  - `learning-systems` - Confidence decay, pattern maturity
-  - `skill-creator` - Meta-skill for authoring new skills
-  - `system-design` - Architecture decision frameworks
-
-  ### Why It Matters
-
-  **Before:** Two overlapping skill systems causing confusion. Agents could use plugin tools OR OpenCode's native syntax, with different behavior and semantics.
-
-  **After:** One canonical path. OpenCode owns discovery and loading. Plugin owns authoring and validation. Clean separation of concerns.
-
-  **Benefits:**
-
-  - No tool conflicts between plugin and platform
-  - Native OpenCode syntax (`use skill testing-patterns`) works seamlessly
-  - Simpler mental model for users
-  - Authoring tools remain for creating spec-compliant skills
-
-  ### Migration Path
-
-  **For Discovery/Loading:**
-
-  ```typescript
-  // OLD (deprecated, still works but warns)
-  skills_list()
-  skills_use(name="testing-patterns")
-
-  // NEW (OpenCode native syntax)
-  use skill testing-patterns
-  use skill cli-builder with "building argument parser"
-  ```
-
-  **For Authoring (no change needed):**
-
-  ```typescript
-  // Still fully supported
-  skills_create((name = "my-skill"), (description = "Domain expertise"));
-  skills_update((name = "my-skill"), (content = "Updated SKILL.md"));
-  skills_add_script(
-    (skill_name = "my-skill"),
-    (script_name = "validate.ts"),
-    (content = "...")
-  );
-  ```
-
-  ### Backward Compatibility
-
-  **Yes, with warnings.** Deprecated tools continue to function but emit console warnings directing users to OpenCode's native syntax. No breaking changes in this release.
-
-  Future major version (v1.0) will remove deprecated discovery tools entirely. Authoring tools remain permanent.
-
-  ### What This Means for Bundled Skills
-
-  Nothing changes. All 6 global skills ship with the plugin and are accessible via OpenCode's native `use skill <name>` syntax. They follow the Agent Skills spec and work identically whether loaded via deprecated plugin tools or native OpenCode.
-
-  The `global-skills/` directory remains the canonical source for our curated skill library.
-
-- [`4feebaf`](https://github.com/joelhooks/swarm-tools/commit/4feebafed61caa8e2e8729b44bd415d71afd6834) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 LLM-Powered Compaction: The Swarm Remembers
-
-  > "The best way to predict the future is to invent it." — Alan Kay
-
-  Compaction just got smarter. Instead of static "here's what to preserve" instructions, the swarm now **generates dynamic continuation prompts** with actual state data.
-
-  **What changed:**
-
-  The `experimental.session.compacting` hook now uses a three-level fallback chain:
-
-  1. **LLM-Generated Prompt** (NEW) - Queries actual swarm state (cells, epics, subtasks), shells out to `opencode run -m <liteModel>` to generate a structured continuation prompt with real IDs, real status, real next actions
-  2. **Static Context** - Falls back to `SWARM_COMPACTION_CONTEXT` if LLM fails
-  3. **Detection Fallback** - For low-confidence swarm detection, injects `SWARM_DETECTION_FALLBACK`
-  4. **None** - No injection if no swarm evidence
-
-  **Progressive Enhancement:**
-
-  Uses OpenCode PR #5907's new `output.prompt` API when available:
-
-  ```typescript
-  if ("prompt" in output) {
-    output.prompt = llmGeneratedPrompt; // Replaces entire compaction prompt
-  } else {
-    output.context.push(llmGeneratedPrompt); // Old API fallback
-  }
-  ```
-
-  **New interfaces:**
-
-  - `SwarmStateSnapshot` - Structured state for LLM input
-  - `querySwarmState()` - Queries cells via swarm CLI
-  - `generateCompactionPrompt()` - Shells out to lite model (30s timeout)
-
-  **Why it matters:**
-
-  Before: "Hey, you should preserve swarm state" (agent has to figure out what that means)
-  After: "Here's epic bd-xyz with 3/5 subtasks done, bd-xyz.2 is blocked on auth, spawn bd-xyz.4 next"
-
-  The coordinator wakes up from compaction with **concrete data**, not instructions to go find data.
-
-  **Backward compatible:** Falls back gracefully on older OpenCode versions or LLM failures.
-
-- [`652fd16`](https://github.com/joelhooks/swarm-tools/commit/652fd16ff424eff92ebb3f5da0599caf676de2ce) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔭 Observability Stack MVP: See What Your Swarm Is Doing
-
-  > "You can't improve what you can't measure." — Peter Drucker
-
-  The swarm just got eyes. This release adds comprehensive observability for multi-agent coordination, answering the eternal question: "Why did my epic fail?"
-
-  ### What's New
-
-  **Structured Error Classes** (swarm-mail)
-
-  - `BaseSwarmError` with rich context: agent, bead_id, epic_id, timestamp, recent events
-  - Specialized errors: `ReservationError`, `CheckpointError`, `ValidationError`, `DecompositionError`
-  - Every error includes actionable suggestions for resolution
-  - Full `toJSON()` serialization for logging and debugging
-
-  **DEBUG Logging** (swarm-mail)
-
-  - `DEBUG=swarm:*` environment variable filtering
-  - 4 subsystems: `swarm:events`, `swarm:reservations`, `swarm:messages`, `swarm:checkpoints`
-  - Zero overhead when disabled
-
-  **swarm-db CLI** (swarm-mail)
-
-  ```bash
-  # Raw SQL queries (SELECT only, max 1000 rows)
-  swarm-db query "SELECT type, COUNT(*) FROM events GROUP BY type"
-
-  # Pre-built analytics
-  swarm-db analytics failed-decompositions --since 7d --format json
-
-  # List available analytics
-  swarm-db list
-  ```
-
-  **10 Pre-built Analytics Queries** (Four Golden Signals mapped)
-  | Query | What It Answers |
-  |-------|-----------------|
-  | `failed-decompositions` | Which strategies are failing? |
-  | `strategy-success-rates` | What's working? |
-  | `lock-contention` | Where are agents fighting over files? |
-  | `agent-activity` | Who's doing what? |
-  | `message-latency` | How fast is coordination? |
-  | `scope-violations` | Who's touching files they shouldn't? |
-  | `task-duration` | How long do tasks take? |
-  | `checkpoint-frequency` | Are agents checkpointing enough? |
-  | `recovery-success` | Do checkpoints actually help? |
-  | `human-feedback` | What are reviewers rejecting? |
-
-  **Agent-Facing Tools** (opencode-swarm-plugin)
-
-  ```typescript
-  // Query analytics programmatically
-  swarm_analytics({
-    query: "failed-decompositions",
-    since: "7d",
-    format: "summary",
-  });
-
-  // Raw SQL for power users (max 50 rows, context-safe)
-  swarm_query({ sql: "SELECT * FROM events WHERE type = 'task_blocked'" });
-
-  // Auto-diagnosis for debugging
-  swarm_diagnose({
-    epic_id: "bd-123",
-    include: ["blockers", "errors", "timeline"],
-  });
-
-  // Learning insights for feedback loops
-  swarm_insights({ scope: "epic", metrics: ["success_rate", "avg_duration"] });
-  ```
-
-  ### Why This Matters
-
-  Before: "The swarm failed. No idea why."
-  After: "Strategy X failed 80% of the time due to file conflicts. Switching to Y."
-
-  Event sourcing was already 80% of the solution. This release adds the diagnostic views to make that data actionable.
-
-  ### Test Coverage
-
-  - 588 tests passing
-  - 1214 assertions
-  - Full TDD: every feature started with a failing test
-
-- [`ca9936d`](https://github.com/joelhooks/swarm-tools/commit/ca9936d09b749449ef3c88fd3ec8b937f6ed7c29) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔬 Research Phase: Docs Before Decomposition
-
-  Swarm coordinators now gather documentation BEFORE breaking down tasks. No more workers fumbling through outdated API assumptions.
-
-  **What's New:**
-
-  - **swarm/researcher agent** - READ-ONLY doc gatherer that discovers tools, reads lockfiles, fetches version-specific docs, and stores findings in semantic-memory
-  - **Pre-decomposition research** - Coordinator analyzes task → identifies tech stack → spawns researchers → injects findings into shared_context
-  - **On-demand research for workers** - Workers can spawn researchers when hitting unknowns mid-task
-  - **`--check-upgrades` flag** - Compare installed vs latest versions from npm registry
-
-  **New Tools:**
-
-  | Tool                     | Purpose                                                     |
-  | ------------------------ | ----------------------------------------------------------- |
-  | `swarm_discover_tools`   | Runtime discovery of available doc tools (MCP, CLI, skills) |
-  | `swarm_get_versions`     | Parse lockfiles (npm/pnpm/yarn/bun) for installed versions  |
-  | `swarm_spawn_researcher` | Generate researcher prompt for Task tool                    |
-  | `swarm_research_phase`   | Manual trigger for research orchestration                   |
-
-  **Architecture:**
-
-  ```
-  Coordinator receives task
-      ↓
-  runResearchPhase(task, projectPath)
-      ↓
-    extractTechStack() → identify technologies
-    discoverDocTools() → find available tools
-    getInstalledVersions() → read lockfiles
-    Spawn researchers (parallel)
-    Collect summaries → shared_context
-      ↓
-  Normal decomposition with enriched context
-  ```
-
-  **Why This Matters:**
-
-  Workers now start with version-specific documentation instead of hallucinating APIs. Researchers store detailed findings in semantic-memory, so future agents don't repeat the research.
-
-### Patch Changes
-
-- Updated dependencies [[`652fd16`](https://github.com/joelhooks/swarm-tools/commit/652fd16ff424eff92ebb3f5da0599caf676de2ce)]:
-  - swarm-mail@1.4.0
-
-## 0.32.0
-
-### Minor Changes
-
-- [#54](https://github.com/joelhooks/swarm-tools/pull/54) [`358e18f`](https://github.com/joelhooks/swarm-tools/commit/358e18f0f7f18d03492ef16c2c1d3edd85c00101) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🔍 Coordinator Review Gate + UBS Removal
-
-  > _"This asynchronous back and forth between submitter and reviewer can add days to the process of getting changes made. Do Code Reviews Promptly!"_
-  > — Sam Newman, _Building Microservices_
-
-  Two changes that make swarm coordination tighter:
-
-  ### Coordinator Review Tools
-
-  New tools for coordinators to review worker output before approval:
-
-  ```
-  ┌─────────────────────────────────────────────────────┐
-  │              COORDINATOR REVIEW FLOW                │
-  ├─────────────────────────────────────────────────────┤
-  │  1. Worker completes → sends completion message     │
-  │  2. Coordinator: swarm_review(task_id, files)       │
-  │     → Gets diff + epic context + review prompt      │
-  │  3. Coordinator reviews against epic goals          │
-  │  4. swarm_review_feedback(status, issues)           │
-  │     → approved: worker can finalize                 │
-  │     → needs_changes: worker gets feedback           │
-  │  5. 3-strike rule: 3 rejections = blocked           │
-  └─────────────────────────────────────────────────────┘
-  ```
-
-  **New tools:**
-
-  - `swarm_review` - Generate review prompt with epic context + git diff
-  - `swarm_review_feedback` - Send approval/rejection with structured issues
-
-  **Updated prompts:**
-
-  - Coordinator prompt now includes review checklist
-  - Worker prompt explains the review gate
-  - Skills updated with review patterns
-
-  ### UBS Scan Removed from swarm_complete
-
-  The `skip_ubs_scan` parameter is gone. UBS was already disabled in v0.31 for performance - this cleans up the vestigial code.
-
-  **Removed:**
-
-  - `skip_ubs_scan` parameter from schema
-  - `ubs_scan` deprecation object from output
-  - All UBS-related helper functions
-  - ~100 lines of dead code
-
-  **If you need UBS scanning:** Run it manually before commit:
-
-  ```bash
-  ubs scan src/
-  ```
-
-  ### CLI Improvements
-
-  The `swarm` CLI got smarter:
-
-  - Better error messages for missing dependencies
-  - Cleaner output formatting
-  - Improved help text
-
-### Patch Changes
-
-- [#54](https://github.com/joelhooks/swarm-tools/pull/54) [`358e18f`](https://github.com/joelhooks/swarm-tools/commit/358e18f0f7f18d03492ef16c2c1d3edd85c00101) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🧪 Integration Test Coverage: 0% → 95%
-
-  > _"Many characterization tests look like 'sunny day' tests. They don't test many special conditions; they just verify that particular behaviors are present. From their presence, we can infer that refactoring hasn't broken anything."_
-  > — Michael Feathers, _Working Effectively with Legacy Code_
-
-  We had a bug that broke ALL swarm tools:
-
-  ```
-  Error: [streams/store] dbOverride parameter is required for this function.
-  PGlite getDatabase() has been removed.
-  ```
-
-  **Why didn't tests catch it?** No integration tests exercised the full tool → store → DB path.
-
-  **Now they do.**
-
-  ```
-  ┌─────────────────────────────────────────────────────────────────┐
-  │              tool-adapter.integration.test.ts                   │
-  ├─────────────────────────────────────────────────────────────────┤
-  │  20 tests | 75 assertions | 1.3s                                │
-  │                                                                 │
-  │  ✅ swarmmail_* tools (6 tests)                                 │
-  │  ✅ hive_* tools (7 tests)                                      │
-  │  ✅ swarm_progress, swarm_status (2 tests)                      │
-  │  ✅ swarm_broadcast, swarm_checkpoint (2 tests)                 │
-  │  ✅ semantic_memory_store, semantic_memory_find (2 tests)       │
-  │  ✅ Smoke test - 9 tools in sequence (1 test)                   │
-  └─────────────────────────────────────────────────────────────────┘
-  ```
-
-  ### What's Tested
-
-  Each test calls `tool.execute()` and verifies:
-
-  1. No "dbOverride required" error (the bug symptom)
-  2. Tool returns expected structure
-  3. Full path works: tool → store → DB → response
-
-  ### The Smoke Test
-
-  Runs 9 tools in sequence to catch interaction bugs:
-
-  ```
-  swarmmail_init → hive_create → swarmmail_reserve → swarm_progress
-  → semantic_memory_store → semantic_memory_find → swarmmail_send
-  → hive_close → swarmmail_release
-  ```
-
-  If ANY step throws "dbOverride required", the test fails.
-
-  ### Also Fixed
-
-  - **Auto-adapter creation** in store.ts - functions now auto-create adapters when not provided
-  - **Exported `clearAdapterCache()`** for test isolation
-  - **Migrated test files** from old `getDatabase()` to adapter pattern
-
-  ### Mandatory Coordinator Review Loop
-
-  Added `COORDINATOR_POST_WORKER_CHECKLIST` constant and `post_completion_instructions` field to `swarm_spawn_subtask`. Coordinators now get explicit instructions to review worker output before spawning the next worker.
-
-  The "dbOverride required" bug **cannot recur undetected**.
-
-- Updated dependencies [[`358e18f`](https://github.com/joelhooks/swarm-tools/commit/358e18f0f7f18d03492ef16c2c1d3edd85c00101), [`358e18f`](https://github.com/joelhooks/swarm-tools/commit/358e18f0f7f18d03492ef16c2c1d3edd85c00101), [`358e18f`](https://github.com/joelhooks/swarm-tools/commit/358e18f0f7f18d03492ef16c2c1d3edd85c00101), [`358e18f`](https://github.com/joelhooks/swarm-tools/commit/358e18f0f7f18d03492ef16c2c1d3edd85c00101)]:
-  - swarm-mail@1.3.0
-
-## 0.31.7
-
-### Patch Changes
-
-- [`97e89a6`](https://github.com/joelhooks/swarm-tools/commit/97e89a6d944b70f205eeb83eb3f2c55a42f5dc08) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Setup Skips Already-Migrated Memories
-
-  `swarm setup` now detects when semantic memories have already been migrated and skips the migration prompt entirely.
-
-  **Before:** Setup would prompt "Migrate to swarm-mail database?" even when all memories were already migrated, then hang.
-
-  **After:** Setup checks if target database has memories first. If already migrated, shows dim "Already migrated to swarm-mail" and moves on.
-
-  **Changes:**
-
-  - Added `targetHasMemories(targetDb)` function to swarm-mail
-  - Updated setup flow to check target before prompting
-  - Fixed connection cleanup in all code paths (try/finally pattern)
-  - Suppressed internal PGLite NOTICE messages from user output
-
-  **Root cause of hang:** PGLite connection wasn't being closed in all paths, keeping the Node.js event loop alive indefinitely.
-
-- Updated dependencies [[`97e89a6`](https://github.com/joelhooks/swarm-tools/commit/97e89a6d944b70f205eeb83eb3f2c55a42f5dc08)]:
-  - swarm-mail@1.2.2
-
-## 0.31.6
-
-### Patch Changes
-
-- [`3147d36`](https://github.com/joelhooks/swarm-tools/commit/3147d36cf2355b9cfe461c7dfc3b30675ea36d89) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🚪 Setup Now Exits Cleanly After Migration
-
-  Fixed a process hang where `swarm setup` would complete the migration but never exit.
-
-  **Root cause:** The PGLite connection created for memory migration kept the Node.js event loop alive indefinitely.
-
-  **Fix:** Close the swarmMail connection after migration completes. The connection is scoped to the migration block and not needed afterward.
-
-  ```typescript
-  // After migration completes
-  await swarmMail.close();
-  ```
-
-  **Before:** `swarm setup` hangs after "Migration complete" message
-  **After:** Process exits cleanly, returns to shell
-
-## 0.31.5
-
-### Patch Changes
-
-- Updated dependencies [[`64368aa`](https://github.com/joelhooks/swarm-tools/commit/64368aa6106089346cd2b1324f6235d5c673964b)]:
-  - swarm-mail@1.2.1
-
-## 0.31.4
-
-### Patch Changes
-
-- Updated dependencies [[`70ff3e0`](https://github.com/joelhooks/swarm-tools/commit/70ff3e054cd1991154f7631ce078798de1076ba8)]:
-  - swarm-mail@1.2.0
-
-## 0.31.3
-
-### Patch Changes
-
-- [`fdddd27`](https://github.com/joelhooks/swarm-tools/commit/fdddd27f9c8627f7de2b9f108827c66c7040b049) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Short Hashes Now Welcome
-
-  The WorkerHandoff schema was too strict - it rejected short project names and partial hashes.
-
-  **Before:** Required 3+ hyphen-separated segments (regex nightmare)
-
-  ```
-  /^[a-z0-9]+(-[a-z0-9]+){2,}(\.[\w-]+)?$/
-  ```
-
-  **After:** Any non-empty string, validated at runtime via `resolvePartialId()`
-
-  Now you can use:
-
-  - Full IDs: `opencode-swarm-monorepo-lf2p4u-mjd4pjujc7e`
-  - Short hashes: `mjd4pjujc7e`
-  - Partial hashes: `mjd4pjuj`
-
-  The hive tools already had smart ID resolution - we just needed to stop blocking it at the schema level.
-
-## 0.31.2
-
-### Patch Changes
-
-- [`d5ec86e`](https://github.com/joelhooks/swarm-tools/commit/d5ec86e77bdb1cd06cf168946aaaff91208dfac1) Thanks [@joelhooks](https://github.com/joelhooks)! - Rebuild with fixed swarm-mail dependency (bigint date fix)
-
-## 0.31.1
-
-### Patch Changes
-
-- [`19995a6`](https://github.com/joelhooks/swarm-tools/commit/19995a68dd1283de1d13afa6fc028bd1273d1b27) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Squashed the BigInt Date Bug
-
-  PGLite returns BIGINT columns as JavaScript `bigint` type. The `Date` constructor throws when given a bigint:
-
-  ```javascript
-  new Date(1734628445371n); // TypeError: Cannot convert a BigInt value to a number
-  ```
-
-  This caused `Invalid Date` errors in all hive operations (`hive_query`, `hive_create`, etc).
-
-  **Fix:** Wrap timestamps in `Number()` before passing to `Date`:
-
-  ```typescript
-  // Before (broken)
-  new Date(cell.created_at);
-
-  // After (works with both number and bigint)
-  new Date(Number(cell.created_at));
-  ```
-
-  **Files fixed:**
-
-  - `swarm-mail/src/hive/jsonl.ts` - JSONL export functions
-  - `opencode-swarm-plugin/src/hive.ts` - `formatCellForOutput()`
-
-  **Tests added:** 6 new tests covering bigint date handling edge cases.
-
-- Updated dependencies [[`19995a6`](https://github.com/joelhooks/swarm-tools/commit/19995a68dd1283de1d13afa6fc028bd1273d1b27)]:
-  - swarm-mail@1.1.1
-
-## 0.31.0
-
-### Minor Changes
-
-- [`39593d7`](https://github.com/joelhooks/swarm-tools/commit/39593d7ee817c683ad1877af52ad5f2ca140c4e2) Thanks [@joelhooks](https://github.com/joelhooks)! - ## Smart ID Resolution: Git-Style Partial Hashes for Hive
-
-  ```
-  ┌─────────────────────────────────────────────────────────────┐
-  │  BEFORE: hive_close(id="opencode-swarm-monorepo-lf2p4u-mjcadqq3fb9")  │
-  │  AFTER:  hive_close(id="mjcadqq3fb9")                                 │
-  └─────────────────────────────────────────────────────────────┘
-  ```
-
-  Cell IDs got long. Now you can use just the hash portion.
-
-  **What changed:**
-
-  ### swarm-mail
-
-  - Added `resolvePartialId(adapter, partialId)` to resolve partial hashes to full cell IDs
-  - Supports exact match, prefix match, suffix match, and substring match
-  - Returns helpful error messages for ambiguous matches ("Found 3 cells matching 'abc': ...")
-  - 36 new tests covering all resolution scenarios
-
-  ### opencode-swarm-plugin
-
-  - `hive_update`, `hive_close`, `hive_start` now accept partial IDs
-  - Resolution happens transparently - full ID returned in response
-  - Backward compatible - full IDs still work
-
-  **JSONL Fix (bonus):**
-
-  - `serializeToJSONL()` now adds trailing newline for POSIX compliance
-  - Prevents parse errors when appending to existing files
-
-  **Why it matters:**
-
-  - Less typing, fewer copy-paste errors
-  - Matches git's partial SHA workflow (muscle memory)
-  - Ambiguous matches fail fast with actionable error messages
-
-  > "The best interface is no interface" - Golden Krishna
-  > (But if you must have one, make it forgive typos)
-
-  ***
-
-  ## Auto-Sync at Key Events
-
-  ```
-  ┌─────────────────────────────────────────┐
-  │  hive_create_epic  →  auto-sync         │
-  │  swarm_complete    →  auto-sync         │
-  │  process.exit      →  safety net sync   │
-  └─────────────────────────────────────────┘
-  ```
-
-  Cells no longer get lost when processes exit unexpectedly.
-
-  **What changed:**
-
-  - `hive_create_epic` syncs after creating epic + subtasks (workers can see them immediately)
-  - `swarm_complete` syncs before worker exits (completed work persists)
-  - `process.on('beforeExit')` hook catches any remaining dirty cells
-
-  **Why it matters:**
-
-  - Spawned workers couldn't see cells created by coordinator (race condition)
-  - Worker crashes could lose completed work
-  - Now the lazy-write pattern has strategic checkpoints
-
-  ***
-
-  ## Removed Arbitrary Subtask Limits
-
-  ```
-  BEFORE: max_subtasks capped at 10 (why tho?)
-  AFTER:  no limit - LLM decides based on task complexity
-  ```
-
-  **What changed:**
-
-  - Removed `.max(10)` from `swarm_decompose` and `swarm_plan_prompt`
-  - `max_subtasks` is now optional with no default
-  - Prompt says "as many as needed" instead of "2-10"
-
-  **Why it matters:**
-
-  - Complex epics need more than 10 subtasks
-  - Arbitrary limits force awkward decomposition
-  - Trust the coordinator to make good decisions
-
-### Patch Changes
-
-- Updated dependencies [[`39593d7`](https://github.com/joelhooks/swarm-tools/commit/39593d7ee817c683ad1877af52ad5f2ca140c4e2)]:
-  - swarm-mail@1.1.0
-
-## 0.30.7
-
-### Patch Changes
-
-- Updated dependencies [[`230e9aa`](https://github.com/joelhooks/swarm-tools/commit/230e9aa91708610183119680cb5f6924c1089552), [`181fdd5`](https://github.com/joelhooks/swarm-tools/commit/181fdd507b957ceb95e069ae71d527d3f7e1b940)]:
-  - swarm-mail@1.0.0
-
-## 0.30.6
-
-### Patch Changes
-
-- [`32a2885`](https://github.com/joelhooks/swarm-tools/commit/32a2885115cc3e574e86d8e492f60ee189627488) Thanks [@joelhooks](https://github.com/joelhooks)! - chore: verify CI publish flow works
-
-## 0.30.5
-
-### Patch Changes
-
-- [`08e61ab`](https://github.com/joelhooks/swarm-tools/commit/08e61abd96ced0443a5ac5dca0e8f362ed869075) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Workers Now Choose Their Own Model
-
-  Added intelligent model selection for swarm workers based on task characteristics.
-
-  **What changed:**
-
-  - `swarm setup` now asks for a "lite model" preference (docs/tests/simple edits)
-  - New `selectWorkerModel()` function auto-selects based on file types
-  - `swarm_spawn_subtask` includes `recommended_model` in metadata
-  - `DecomposedSubtask` schema supports optional explicit `model` field
-
-  **Model selection priority:**
-
-  1. Explicit `model` field in subtask (if specified)
-  2. File-type inference:
-     - All `.md`/`.mdx` files → lite model
-     - All `.test.`/`.spec.` files → lite model
-  3. Mixed or implementation files → primary model
-
-  **Why it matters:**
-
-  - Cost savings: docs and tests don't need expensive models
-  - Faster execution: lite models are snappier for simple tasks
-  - Better defaults: right-sized models for each subtask type
-  - Still flexible: coordinators can override per-subtask
-
-  **Backward compatible:**
-
-  - Existing workflows continue to work
-  - Model selection is transparent to agents
-  - Defaults to primary model if lite model not configured
-
-  **Example:**
-
-  ```typescript
-  // Subtask with all markdown files
-  { files: ["README.md", "docs/guide.mdx"] }
-  // → selects lite model (haiku)
-
-  // Subtask with mixed files
-  { files: ["src/auth.ts", "README.md"] }
-  // → selects primary model (sonnet)
-
-  // Explicit override
-  { files: ["complex-refactor.ts"], model: "anthropic/claude-opus-4-5" }
-  // → uses opus as specified
-  ```
-
-## 0.30.4
-
-### Patch Changes
-
-- [`1c9a2e8`](https://github.com/joelhooks/swarm-tools/commit/1c9a2e8a148b79a33cb8c5b565e485f33d1f617c) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Fix Migration Adapter Type (for real this time)
-
-  The previous release (0.30.3) was published before this fix landed. Now it's actually in.
-
-  **The Bug:**
-
-  ```
-  targetDb.query is not a function
-  ```
-
-  **Root Cause:**
-  `getSwarmMail()` returns `SwarmMailAdapter`, not `DatabaseAdapter`. Need to call `getDatabase()` first.
-
-  **The Fix:**
-
-  ```typescript
-  const swarmMail = await getSwarmMail(cwd);
-  const targetDb = await swarmMail.getDatabase(cwd);
-  ```
-
-## 0.30.3
-
-### Patch Changes
-
-- [`cc84c8f`](https://github.com/joelhooks/swarm-tools/commit/cc84c8f066696c7625dc307a5163ff50d672634e) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Fix Migration Adapter Type Mismatch
-
-  > _"The compiler is your friend. Listen to it."_
-  > — Every TypeScript developer, eventually
-
-  Fixed a runtime error in `swarm setup` where the legacy memory migration was receiving a `SwarmMailAdapter` instead of a `DatabaseAdapter`.
-
-  **The Bug:**
-
-  ```
-  targetDb.query is not a function
-  ```
-
-  **Root Cause:**
-  `getSwarmMail()` returns a `SwarmMailAdapter` which has `getDatabase()` method, not a direct `query()` method. The migration code expected a `DatabaseAdapter`.
-
-  **The Fix:**
-
-  ```typescript
-  // Before (wrong)
-  const targetDb = await getSwarmMail(cwd);
-
-  // After (correct)
-  const swarmMail = await getSwarmMail(cwd);
-  const targetDb = await swarmMail.getDatabase(cwd);
-  ```
-
-  **Test Added:**
-  New test case verifies that passing an invalid adapter (without `query()`) fails gracefully with a descriptive error instead of crashing.
-
-- [`1e41c9b`](https://github.com/joelhooks/swarm-tools/commit/1e41c9b42ae468761f813d406171d182fb9948e0) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Semantic Memory Consolidation
-
-  > _"Simplicity is the ultimate sophistication."_
-  > — Leonardo da Vinci
-
-  The semantic memory system has moved into swarm-mail, bringing persistent learning to the hive.
-
-  ### What's New
-
-  **Semantic Memory in swarm-mail:**
-
-  - `createSemanticMemory()` - Initialize memory store with PGLite + Ollama embeddings
-  - `getMigrationStatus()` - Check if legacy memory needs migration
-  - `migrateLegacyMemory()` - Migrate from old semantic-memory-mcp format
-  - Automatic migration on first use (no manual intervention needed)
-
-  **Legacy Migration:**
-
-  - Detects old `~/.semantic-memory/` databases
-  - Migrates memories, embeddings, and metadata
-  - Preserves all tags and timestamps
-  - Creates backup before migration
-
-  **Worker Handoff Protocol:**
-
-  - Agents can now hand off work mid-task
-  - State preserved via swarm mail messages
-  - Enables long-running tasks across context limits
-
-  ### Breaking Changes
-
-  None - this is additive. The old semantic-memory-mcp still works but is deprecated.
-
-  ### Files Added/Changed
-
-  - `packages/swarm-mail/src/memory/` - New memory subsystem
-  - `packages/swarm-mail/src/memory/migrate-legacy.ts` - Migration tooling
-  - `packages/opencode-swarm-plugin/bin/swarm.ts` - Uses new exports
-
-- Updated dependencies [[`1e41c9b`](https://github.com/joelhooks/swarm-tools/commit/1e41c9b42ae468761f813d406171d182fb9948e0)]:
-  - swarm-mail@0.5.0
-
-## 0.30.2
-
-### Patch Changes
-
-- [`5858148`](https://github.com/joelhooks/swarm-tools/commit/5858148d5785393c0a6993a2595fba275f305707) Thanks [@joelhooks](https://github.com/joelhooks)! - chore: trigger publish workflow
-
-## 0.30.1
-
-### Patch Changes
-
-- [`57d5600`](https://github.com/joelhooks/swarm-tools/commit/57d5600a53e148ce1d1da48b3b5a8723a5552e04) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🚦 Review Gate UX Fix + Verbose Setup
-
-  > _"A common mistake that people make when trying to design something completely foolproof is to underestimate the ingenuity of complete fools."_
-  > — Douglas Adams, _Mostly Harmless_
-
-  Two UX improvements that make swarm coordination feel less like shouting into the void.
-
-  ### What Changed
-
-  **Review Gate Response Fix:**
-
-  - `swarm_complete` no longer returns `success: false` when code review is pending
-  - Now returns `success: true` with `status: "pending_review"` or `status: "needs_changes"`
-  - **Why it matters**: The old format made review checkpoints look like errors. Agents would retry unnecessarily or report failures when the workflow was actually working as designed. Review gates are a feature, not a bug.
-
-  **Setup Command Verbosity:**
-
-  - Added `p.log.step()` and `p.log.success()` throughout swarm setup
-  - Users can now see exactly what's happening: dependency checks, git init, swarm-mail connection
-  - **Why it matters**: Silent setup commands feel broken. Explicit progress logs build trust and make debugging easier when setup actually does fail.
-
-  ### Why It Matters
-
-  **For Agents:**
-
-  - No more false-negative responses from review gates
-  - Clear workflow state (pending vs. needs changes vs. complete)
-  - Reduced retry loops and error noise
-
-  **For Users:**
-
-  - Setup command shows its work (not a black box)
-  - Review process is transparent in logs
-  - Easier to diagnose when things actually break
-
-  **Backward compatible:** Yes. Existing agents checking for `success: false` will still work, they just won't see false errors anymore.
-
-## 0.30.0
-
-### Minor Changes
-
-- [`f3917ad`](https://github.com/joelhooks/swarm-tools/commit/f3917ad911d3c716a2470a01c66bce3500f644f4) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 The Great bd CLI Purge
-
-  The `bd` CLI is officially dead. Long live HiveAdapter!
-
-  **What changed:**
-
-  ### `swarm init` Command Rewritten
-
-  - No longer shells out to `bd init` or `bd create`
-  - Uses `ensureHiveDirectory()` and `getHiveAdapter()` directly
-  - Supports `.beads` → `.hive` migration with user prompts
-  - Creates cells via HiveAdapter, not CLI
-
-  ### Auto-sync Removed from `index.ts`
-
-  - Removed `void $\`bd sync\`.quiet().nothrow()`after`hive_close`
-  - Users should call `hive_sync` explicitly at session end
-  - This was a fire-and-forget that could race with other operations
-
-  ### Plugin Template Updated
-
-  - `detectSwarm()` now has confidence levels (HIGH/MEDIUM/LOW/NONE)
-  - Added `SWARM_DETECTION_FALLBACK` for uncertain cases
-  - Compaction hook injects context based on confidence:
-    - HIGH/MEDIUM → Full swarm context
-    - LOW → Fallback detection prompt
-    - NONE → No injection
-
-  ### Error Handling Fixed
-
-  - `execTool()` now handles both string and object error formats
-  - Fixes "Tool execution failed" generic error from `swarm_complete`
-  - Actual error messages now propagate to the agent
-
-  **Why it matters:**
-
-  - No external CLI dependency for core functionality
-  - HiveAdapter is type-safe and testable
-  - Plugin works in environments without `bd` installed
-  - Better error messages for debugging
-
-  **Migration:** Run `swarm setup` to update your deployed plugin.
-
-## 0.29.0
-
-### Minor Changes
-
-- [`a2ff1f4`](https://github.com/joelhooks/swarm-tools/commit/a2ff1f4257a2e9857f63abe4e9b941a573f44380) Thanks [@joelhooks](https://github.com/joelhooks)! - ## 🐝 Cell IDs Now Wear Their Project Colors
-
-  > _"We may fantasize about being International Men of Mystery, but our code needs to be mundane and clear. One of the most important parts of clear code is good names."_
-  > — Martin Fowler, _Refactoring_
-
-  Cell IDs finally know where they came from. Instead of anonymous `bd-xxx` prefixes,
-  new cells proudly display their project name: `swarm-mail-lf2p4u-abc123`.
-
-  ### What Changed
-
-  **swarm-mail:**
-
-  - `generateBeadId()` now reads `package.json` name field from project directory
-  - Added `slugifyProjectName()` for safe ID generation (lowercase, special chars → dashes)
-  - Falls back to `cell-` prefix if no package.json or no name field
-
-  **opencode-swarm-plugin:**
-
-  - Removed all `bd` CLI usage from `swarm-orchestrate.ts` - now uses HiveAdapter
-  - Improved compaction hook swarm detection with confidence levels (high/medium/low)
-  - Added fallback detection prompt for uncertain swarm states
-
-  ### Examples
-
-  | Before                  | After                           |
-  | ----------------------- | ------------------------------- |
-  | `bd-lf2p4u-mjbneh7mqah` | `swarm-mail-lf2p4u-mjbneh7mqah` |
-  | `bd-abc123-xyz`         | `my-cool-app-abc123-xyz`        |
-  | (no package.json)       | `cell-abc123-xyz`               |
-
-  ### Why It Matters
-
-  - **Identifiable at a glance** - Know which project a cell belongs to without looking it up
-  - **Multi-project workspaces** - Filter/search cells by project prefix
-  - **Terminology cleanup** - Removes legacy "bead" (`bd-`) from user-facing IDs
-
-  ### Backward Compatible
-
-  Existing `bd-*` IDs still work fine. No migration needed - only NEW cells get project prefixes.
-
-  ### Compaction: Keeping the Swarm Alive
-
-  > _"Intelligent and structured group dynamics that emerge not from a leader, but from the local interactions of the elements themselves."_
-  > — Daniel Shiffman, _The Nature of Code_
-
-  The compaction hook now uses multi-signal detection to keep swarms cooking through context compression:
-
-  - **HIGH confidence:** Active reservations, in_progress cells → full swarm context
-  - **MEDIUM confidence:** Open subtasks, unclosed epics → full swarm context
-  - **LOW confidence:** Any cells exist → fallback detection prompt
-
-  Philosophy: Err on the side of continuation. A false positive costs context space. A false negative loses the swarm.
-
-### Patch Changes
-
-- Updated dependencies [[`a2ff1f4`](https://github.com/joelhooks/swarm-tools/commit/a2ff1f4257a2e9857f63abe4e9b941a573f44380)]:
-  - swarm-mail@0.4.0
-
-## 0.28.2
-
-### Patch Changes
-
-- Updated dependencies [[`90409ef`](https://github.com/joelhooks/swarm-tools/commit/90409ef4f353844b25fe04221bc80d6f930eced2)]:
-  - swarm-mail@0.3.4
-
-## 0.28.1
-
-### Patch Changes
-
-- [`0ee4f65`](https://github.com/joelhooks/swarm-tools/commit/0ee4f656c2fb2cf62d3ef06d329d9e093d124c33) Thanks [@joelhooks](https://github.com/joelhooks)! - Add postinstall hint and update repo URL
-
-  - Show "Run swarm setup" hint after npm install
-  - Update repo URL to github.com/joelhooks/swarm-tools
-  - Add "Get started" commands to version output
-
-## 0.28.0
-
-### Minor Changes
-
-- [`de2fa62`](https://github.com/joelhooks/swarm-tools/commit/de2fa628524b88511e06164104ff7b5fb93d39e5) Thanks [@joelhooks](https://github.com/joelhooks)! - Add full beads→hive migration pipeline with JSONL import to PGLite
-
-  - Add `mergeHistoricBeads()` to merge beads.base.jsonl into issues.jsonl
-  - Add `importJsonlToPGLite()` to import JSONL records into PGLite database
-  - Wire both functions into `swarm setup` migration flow
-  - Fix closed_at constraint issue when importing closed cells
-  - TDD: 12 new integration tests for migration functions
-
-## 0.27.4
-
-### Patch Changes
-
-- [`f23f774`](https://github.com/joelhooks/swarm-tools/commit/f23f774e4b83a3422d8266b6b1ad083daaec03e2) Thanks [@joelhooks](https://github.com/joelhooks)! - Enforce coordinator always spawns workers, never executes work directly
-
-  - Added "Coordinator Role Boundaries" section to /swarm command
-  - Coordinators now explicitly forbidden from editing code, running tests, or making "quick fixes"
-  - Updated Phase 5 to clarify coordinators NEVER reserve files (workers do)
-  - Updated Phase 6 with patterns for both parallel and sequential worker spawning
-  - Worker agent template now confirms it was spawned correctly and to report coordinator violations
-
-## 0.27.3
-
-### Patch Changes
-
-- [`ec23d25`](https://github.com/joelhooks/swarm-tools/commit/ec23d25aeca667c0294a6255fecf11dd7d7fd6b3) Thanks [@joelhooks](https://github.com/joelhooks)! - Add .beads → .hive directory migration support
-
-  - Fix migration version collision: beadsMigration now v7, cellsViewMigration now v8 (was conflicting with streams v6)
-  - Add `checkBeadsMigrationNeeded()` to detect legacy .beads directories
-  - Add `migrateBeadsToHive()` to rename .beads to .hive
-  - Add `ensureHiveDirectory()` to create .hive if missing (called by hive_sync)
-  - Update hive_sync to ensure .hive directory exists before writing
-  - Add migration prompt to `swarm setup` CLI flow
-
-- Updated dependencies [[`ec23d25`](https://github.com/joelhooks/swarm-tools/commit/ec23d25aeca667c0294a6255fecf11dd7d7fd6b3)]:
-  - swarm-mail@0.3.3
-
-## 0.27.2
-
-### Patch Changes
-
-- [`50a2bf5`](https://github.com/joelhooks/swarm-tools/commit/50a2bf51c5320c038f202191d7acbfd2179f2cb3) Thanks [@joelhooks](https://github.com/joelhooks)! - Fix cells view migration not being applied
-
-  The v7 migration (cellsViewMigration) that creates the `cells` view was added after
-  swarm-mail@0.3.0 was published. This caused `hive_sync` to fail with
-  "relation cells does not exist" because the JSONL export queries the `cells` view.
-
-  This patch ensures the v7 migration is included in the published package.
-
-- Updated dependencies [[`50a2bf5`](https://github.com/joelhooks/swarm-tools/commit/50a2bf51c5320c038f202191d7acbfd2179f2cb3)]:
-  - swarm-mail@0.3.2
-
-## 0.27.0
-
-### Minor Changes
-
-- [`26fd2ef`](https://github.com/joelhooks/swarm-tools/commit/26fd2ef27562edc39f7db7a9cdbed399a465200d) Thanks [@joelhooks](https://github.com/joelhooks)! - Rename beads → hive across the codebase
-
-  - `createBeadsAdapter` → `createHiveAdapter` (old name still exported as alias)
-  - `BeadsAdapter` type → `HiveAdapter` type
-  - All internal references updated to use hive terminology
-  - Backward compatible: old exports still work but are deprecated
-
-- [`ab23071`](https://github.com/joelhooks/swarm-tools/commit/ab23071cc7509c4fc37e1cac0f38a3812022cdf5) Thanks [@joelhooks](https://github.com/joelhooks)! - Add swarm-aware compaction hook to keep swarms cooking after context compression
-
-  - New `experimental.session.compacting` hook detects active swarms and injects recovery context
-  - `hasSwarmSign()` checks for swarm evidence: in-progress beads, subtasks, unclosed epics
-  - Compaction prompt instructs coordinator to immediately resume orchestration
-  - Fix @types/node conflicts by pinning to 22.19.3 in root overrides
-
-### Patch Changes
-
-- Updated dependencies [[`26fd2ef`](https://github.com/joelhooks/swarm-tools/commit/26fd2ef27562edc39f7db7a9cdbed399a465200d)]:
-  - swarm-mail@0.3.0
-
-## 0.26.1
-
-### Patch Changes
-
-- [`b2d4a84`](https://github.com/joelhooks/swarm-tools/commit/b2d4a84748cdef4b9dbca7666dd3d313b6cd2b24) Thanks [@joelhooks](https://github.com/joelhooks)! - Add automatic JSONL migration for beads on first use
-
-  - Auto-migrate from `.beads/issues.jsonl` when database is empty
-  - Fix import to handle missing dependencies/labels/comments arrays
-  - Fix closed bead import to satisfy check constraint (status + closed_at)
-  - Migrates 500+ historical beads seamlessly on first adapter initialization
-
-- Updated dependencies [[`b2d4a84`](https://github.com/joelhooks/swarm-tools/commit/b2d4a84748cdef4b9dbca7666dd3d313b6cd2b24)]:
-  - swarm-mail@0.2.1
-
-## 0.26.0
-
-### Minor Changes
-
-- [`1a7b02f`](https://github.com/joelhooks/swarm-tools/commit/1a7b02f707a1490f14465467c6024331d5064878) Thanks [@joelhooks](https://github.com/joelhooks)! - Add PGLite socket server adapter with hybrid daemon management and move streams storage to $TMPDIR.
-
-  **Socket Server Adapter:**
-
-  - New `createSocketAdapter()` wrapping postgres.js for DatabaseAdapter interface
-  - Daemon lifecycle: `startDaemon()`, `stopDaemon()`, `isDaemonRunning()`, `healthCheck()`
-  - Auto-start daemon on first use with `SWARM_MAIL_SOCKET=true` env var
-  - Graceful fallback to embedded PGLite on failure
-  - CLI: `swarm-mail-daemon start|stop|status`
-
-  **$TMPDIR Storage (BREAKING):**
-
-  - Streams now stored in `$TMPDIR/opencode-<project-name>-<hash>/streams`
-  - Eliminates git pollution from `.opencode/streams/`
-  - Auto-cleaned on reboot (ephemeral coordination state)
-  - New exports: `getProjectTempDirName()`, `hashProjectPath()`
-
-  This fixes the multi-agent PGLite corruption issue by having all agents connect to a single pglite-server daemon via PostgreSQL wire protocol.
-
-### Patch Changes
-
-- Updated dependencies [[`1a7b02f`](https://github.com/joelhooks/swarm-tools/commit/1a7b02f707a1490f14465467c6024331d5064878)]:
-  - swarm-mail@0.2.0
-
-## 0.25.3
-
-### Patch Changes
-
-- [`7471fd4`](https://github.com/joelhooks/swarm-tools/commit/7471fd43ef9b16b32e503d7cd4bdc5b7a74537e4) Thanks [@joelhooks](https://github.com/joelhooks)! - Fix swarm_complete tool execution failures and remove debug logging
-
-  **opencode-swarm-plugin:**
-
-  - Fix: Made sendSwarmMessage non-fatal in swarm_complete - failures no longer cause "Tool execution failed" errors
-  - Fix: Added message_sent and message_error fields to swarm_complete response for better error visibility
-  - Chore: Removed console.log statements from index.ts, swarm-orchestrate.ts, storage.ts, rate-limiter.ts
-  - Test: Added integration tests for swarm_complete error handling
-
-  **swarm-mail:**
-
-  - Chore: Cleaned up debug logging and improved migration handling
-
-- Updated dependencies [[`7471fd4`](https://github.com/joelhooks/swarm-tools/commit/7471fd43ef9b16b32e503d7cd4bdc5b7a74537e4)]:
-  - swarm-mail@0.1.4
-
-## 0.25.2
-
-### Patch Changes
-
-- [`34a2c3a`](https://github.com/joelhooks/swarm-tools/commit/34a2c3a07f036297db449414ef8dbeb7b39721e2) Thanks [@joelhooks](https://github.com/joelhooks)! - Grant swarm workers autonomy to file beads against the epic
-
-  Workers can now create bugs, tech debt, and follow-up tasks linked to their parent epic via `parent_id`. Prompt explicitly encourages workers to file issues rather than silently ignoring them.
-
-## 0.25.1
-
-### Patch Changes
-
-- [`757f4a6`](https://github.com/joelhooks/swarm-tools/commit/757f4a690721b3f04a414e4c1694660862504e54) Thanks [@joelhooks](https://github.com/joelhooks)! - Fix skills_update tool - add `content` parameter as primary (with `body` as backwards-compat alias)
-
-  The tool was only accepting `body` but users expected `content`. Now both work:
-
-  - `skills_update(name="foo", content="new stuff")` - preferred
-  - `skills_update(name="foo", body="new stuff")` - still works for backwards compat
-
-- [`3d619ff`](https://github.com/joelhooks/swarm-tools/commit/3d619ffda78b2e6066491f053e8fad8dac7b5b71) Thanks [@joelhooks](https://github.com/joelhooks)! - Fix swarm_complete failing when bead project doesn't match CWD
-
-  - Use `project_key` as working directory for `bd close` command
-  - Improved error messages with context-specific recovery steps
-  - Added planning guardrails to warn when todowrite is used for parallel work (should use swarm)
-
-## 0.25.0
-
-### Minor Changes
-
-- [`b70ae35`](https://github.com/joelhooks/swarm-tools/commit/b70ae352876515bdfe68511d72bb472c85b7fdfc) Thanks [@joelhooks](https://github.com/joelhooks)! - Add Socratic planning phase and improved worker prompts to swarm setup
-
-  **SWARM_COMMAND template:**
-
-  - Added Phase 0: Socratic Planning - asks clarifying questions before decomposing
-  - Supports `--fast`, `--auto`, `--confirm-only` flags to skip questions
-  - ONE question at a time with concrete options and recommendations
-
-  **Worker agent template:**
-
-  - Reinforces the 9-step survival checklist from SUBTASK_PROMPT_V2
-  - Explicitly lists all steps with emphasis on non-negotiables
-  - Explains WHY skipping steps causes problems (lost work, conflicts, etc.)
-
-  **Agent path consolidation:**
-
-  - Now creates nested paths: `~/.config/opencode/agent/swarm/worker.md`
-  - Matches `Task(subagent_type="swarm/worker")` format
-  - Cleans up legacy flat files (`swarm-worker.md`) on reinstall
-
-  To get the new prompts, run `swarm setup` and choose "Reinstall everything".
-
-## 0.24.0
-
-### Minor Changes
-
-- [`434f48f`](https://github.com/joelhooks/swarm-tools/commit/434f48f207c3509f6b924caeb47cd6e019dcc0e1) Thanks [@joelhooks](https://github.com/joelhooks)! - Add worker survival checklist and Socratic planning for swarm coordination
-
-  **Worker Survival Checklist (9-step mandatory flow):**
-
-  - Workers now follow a strict initialization sequence: swarmmail_init → semantic-memory_find → skills_use → swarmmail_reserve
-  - Workers reserve their own files (coordinators no longer reserve on behalf of workers)
-  - Auto-checkpoint at 25/50/75% progress milestones
-  - Workers store learnings via semantic-memory before completing
-
-  **Socratic Planning:**
-
-  - New `swarm_plan_interactive` tool with 4 modes: socratic (default), fast, auto, confirm-only
-  - Default mode asks clarifying questions before decomposition
-  - Escape hatches for experienced users: `--fast`, `--auto`, `--confirm-only` flags on /swarm command
-
-  **Updated Skills:**
-
-  - swarm-coordination skill now documents worker survival patterns and coordinator rules
-
-### Patch Changes
-
-- [#15](https://github.com/joelhooks/swarm-tools/pull/15) [`299f2d3`](https://github.com/joelhooks/swarm-tools/commit/299f2d3305796bcb411f9b90715cda3513d17b54) Thanks [@tayiorbeii](https://github.com/tayiorbeii)! - Sync bundled skills into the global skills directory during `swarm setup` reinstall, fix bundled-skill path resolution, and make AGENTS.md skill-awareness updates work without relying on `opencode run`.
-
-## 0.23.6
-
-### Patch Changes
-
-- Updated dependencies [[`22befbf`](https://github.com/joelhooks/opencode-swarm-plugin/commit/22befbfa120a37a585cfec0709597172efda92a4)]:
-  - swarm-mail@0.1.3
-
-## 0.23.5
-
-### Patch Changes
-
-- [`3826c6d`](https://github.com/joelhooks/opencode-swarm-plugin/commit/3826c6d887f937ccb201b7c4322cbc7b46823658) Thanks [@joelhooks](https://github.com/joelhooks)! - Fix workspace:\* resolution by running bun install before pack
-
-  The lockfile was stale, causing bun pack to resolve workspace:\* to old versions.
-  Now runs bun install first to ensure lockfile matches current package.json versions.
-
-## 0.23.4
-
-### Patch Changes
-
-- Updated dependencies [[`2d0fe9f`](https://github.com/joelhooks/opencode-swarm-plugin/commit/2d0fe9fc6278874ea6c4a92f0395cbdd11c4e994)]:
-  - swarm-mail@0.1.2
-
-## 0.23.3
-
-### Patch Changes
-
-- [`9c4e4f9`](https://github.com/joelhooks/opencode-swarm-plugin/commit/9c4e4f9511672ab8598c7202850c87acf1bfd4b7) Thanks [@joelhooks](https://github.com/joelhooks)! - Fix swarm-mail package to include dist folder
-
-  - Add files field to swarm-mail package.json to explicitly include dist/
-  - Previous publish was missing build output, causing "Cannot find module" errors
-
-- Updated dependencies [[`9c4e4f9`](https://github.com/joelhooks/opencode-swarm-plugin/commit/9c4e4f9511672ab8598c7202850c87acf1bfd4b7)]:
-  - swarm-mail@0.1.1
-
-## 0.23.2
-
-### Patch Changes
-
-- [`7f9ead6`](https://github.com/joelhooks/opencode-swarm-plugin/commit/7f9ead65dab1dd5dc9aff57df0871cc390556fe1) Thanks [@joelhooks](https://github.com/joelhooks)! - Fix workspace:\* protocol resolution using bun pack + npm publish
-
-  Uses bun pack to create tarball (which resolves workspace:\* to actual versions) then npm publish for OIDC trusted publisher support.
-
-## 0.23.1
-
-### Patch Changes
-
-- [`64ad0e4`](https://github.com/joelhooks/opencode-swarm-plugin/commit/64ad0e4fc033597027e3b0614865cfbf955b5983) Thanks [@joelhooks](https://github.com/joelhooks)! - Fix workspace:\* protocol resolution in npm publish
-
-  Use bun publish instead of npm publish to properly resolve workspace:\* protocols to actual versions.
-
-## 0.23.0
-
-### Minor Changes
-
-- [`b66d77e`](https://github.com/joelhooks/opencode-swarm-plugin/commit/b66d77e484e9b7021b3264d1a7e8f54a16ea5204) Thanks [@joelhooks](https://github.com/joelhooks)! - Add changesets workflow and semantic memory test isolation
-
-  - OIDC publish workflow with GitHub Actions
-  - Changesets for independent package versioning
-  - TEST_SEMANTIC_MEMORY_COLLECTION env var for test isolation
-  - Prevents test pollution of production semantic-memory