🐝 v0.61.0 - prompts now know what time it is and remember what you talked about. Semantic memory recall auto-injects relevant hivemind context. SKOS taxonomy extraction for entity relationships you can actually query.

🐝 v0.60.0 - close tasks with results, link commits to completions, run `swarm doctor --deep --fix` to heal your hive. Status dashboard shows who's blocked on what. Trees now tell the truth.

🐝 v0.59.5 - hivemind_find no longer explodes when Claude sends empty queries. Defense in depth: friendly errors at the tool, fail fast at the adapter, graceful degradation at the store. Your swarm's memory just got less fragile.

🐝 v0.59.0 - swarm stats now shows real success rates. Outcomes link back to decision traces, quality_score actually populates. Your feedback loops were lying to you. Now they're not.

🐝 v0.59.0 - swarm stats no longer lie. Outcomes now link back to decision traces, so quality_score actually populates. Your 0% success rate was a bug, not a judgment.

🐝 v0.59.0 - swarm stats finally shows real success rates. Outcomes now link back to decision traces so quality_score actually calculates. Your 0% wasn't a skill issue, it was a wiring issue.

🐝 v0.58.4 - tools now tell you what's wrong instead of silently failing. Missing params get helpful hints. Subprocesses clean up properly. No more hanging semantic memory calls.

🐝 v0.45.7 - CLI unicode glitches fixed. No more â–¡â–ˆ garbage, actual box characters. Skills directory auto-migrates from plural to singular. Logger works in global installs without worker thread crashes.

🐝 v0.45.6 - swarm setup auto-migrates ~/.config/opencode/skills to skill (singular). No manual work needed.

🐝 v0.45.5 - logs actually work in global installs. Set SWARM_LOG_FILE=1 for file output or get clean stdout JSON. No more worker thread crashes hunting missing modules.

🐝 v0.45.4 - bundled CLI logs actually work now. pino-roll and pino-pretty no longer vanish in webpack. Plus: smart operations migrated from evalite purgatory to bun:test paradise.

🐝 v0.45.3 - fix "Cannot find module evalite/runner" errors during setup. CLI works after npm install. Dumb tools, smart results.

🐝 v0.45.2 - smart operations tests migrated from evalite to bun:test. No more SQLITE_CORRUPT_VTAB pain. LLM memory ops actually test now.

🐝 v0.44.1 - evals extracted to @swarmtools/evals. No more 20MB+ test frameworks in your main plugin. Global installs work again.

🐝 v0.44.0 - contributor_lookup tool fetches GitHub profiles and generates changeset credit lines. Call with login + issue, get formatted thanks ready to paste. No more forgotten contributors.

🐝 v0.43.0 - query your swarm with SQL, replay epic events at 2x speed, export to OTLP traces. Debug coordination failures by watching sequences unfold. 225 tests, TDD enforced.

🐝 v0.42.9 - git worktrees share database state, semantic-memory falls back to full-text when Ollama's down, gh-issue-triage skill for contributor workflows. Fault tolerance > fault elimination.

a quick demo of @swarmtoolsai in action - working on observability, evals, and triaging some gh issues in 3 separate @opencode sessions

🐝 v0.42.8 - release tweets finally link to the actual PR that shipped them. No more stale links, no more wrong commits.

🐝 v0.42.7 - README shows swarm learning architecture in ASCII. See how tasks flow through insights layer, token budgets, API examples you can copy-paste. Diagrams > prose for feedback loops.