Releases

• +Receipts ledger: every ack is a timestamped, per-recipient receipt — who saw this, who approved it, prove it
• +New receipt tool (non-consuming annotations: seen, r-ack, retracted) + get_receipts (full audit trail per message)
• +Fleet broadcast: to_agent_id: "*" delivers to every agent; acknowledged only when all recipients have acked
• +Ledger schema v2 with automatic migration — v1 acks backfilled as receipts, no audit gap

Design ported from a production fleet bus: 10+ agents, 40 days, 18,404 messages. 192 tests across 18 files.

• +Contributor Covenant v2.1 as the project code of conduct
• +Every open source project should have a CoC so contributors know what behavior is expected and how to report violations

Add CODE_OF_CONDUCT.md to the repo. Wire a Conduct link into the homepage footer.

• +Live TUI for running fleets, recent agents (with retry counts), and recent events
• +Refreshes every 1s; ANSI in-place updates; Ctrl+C to exit
• +Verified against a live ledger with 13 fleets

npx agent-mesh-dashboard (or npx agent-mesh dashboard). Closes the last open issue.

• +exportFleetTemplate writes portable meshfleet-template-v1 JSON
• +importFleetTemplate reads, validates the schema, and inserts (auto-suffix on conflict)

Closes issue #7. Templates can now move between machines, back up, or share via Git.

• +saveFleetTemplate now creates a new version on each save (auto-increment)
• +getFleetTemplate(name, version?) returns latest or specific
• +listFleetTemplateVersions + deleteFleetTemplate(name, version?) round out the API

Closes issue #6. Internal ledger key uses `${name}@v${version}` so versions coexist.

• +Self-contained benchmark/bench.ts + BENCHMARKS.md
• +v1.0 perf gates verified: spawn bookkeeping 5.5ms p50 (target <100ms)
• +10k messages persisted with no drops

Known bottleneck: sendMessage rewrites full ledger (3.8ms/msg at 10k scale). Batch writes are v0.9+.

• +Curated synonym table for 30+ common dev terms (frontend ↔ ui/ux/web, database ↔ db/sql, etc.)
• +routeWork now expands 'ui' to also match a frontend agent
• +Zero network cost, no ML dep, override via setSynonymOverrides()

Closes issue #1. Lightweight semantic routing without an embedding model.

• +New record_routing_outcome(agent_id, capability_key, success) MCP tool
• +Wilson-style score adjustment on future route_work calls
• +RouteMatch.weight field exposes the adjusted score

Closes issue #4. Future routing calls weight by past outcomes.

• +Hierarchical skill matching with ancestor/descendant decay
• +JSON taxonomy: { frontend: { react: [nextjs, remix] } }

Closes issue #2. Wire into route_work planned for v0.9.

• +Automatic retry with exponential backoff (1s/2s/4s + ±20% jitter, max 3 attempts)
• +Partial result recovery on startup (interrupted status)
• +Ledger schema versioning (schema_version: 1, auto-migrate v0)
• +route_work top_n parameter for fan-out routing

Closes issues #3, #5. v0.8 series hardening foundation.

• +subscribe_inbox MCP tool (SSE)
• +Standalone HTTP SSE server with per-agent cap + heartbeat
• +notifySubscribers hooked into send_message
• +Heartbeat watchdog (auto-fail agents that miss N liveness checks)

v0.7.x hardening series foundation.