Skip to main content

Hive Activation 2026-04-17 — Main Runbook

Hive Activation — 2026-04-17

Status: Phase 1–5 builders complete; Phase 6 validation in progress. Plan: ~/system/specs/hive-activation-plan.md Evidence: ~/system/evidence/hive-activation-2026-04-17/ Prior sprint: System Evolution 2026-04-16 (see system-evolution-2026-04-16.md).

Why this sprint

After System Evolution we knew:

  • Hivemind grew (31k intel) but subscriptions = 0 — nobody listened.
  • Library had 76 skills but last sync was 26 days old; FORGE never synced.
  • skill-registry.use_count = 0 everywhere — we couldn't tell which skills were alive.
  • discover.js "drop" returned 0 hits in tools/skills/agents/MCP/BookStack.
  • Meta-agent existed on disk (~/.claude/agents/meta-agent.md) but had never produced a proposal.
  • John authored 90% of hivemind writes — the "colony" was a megaphone.

Hive Activation is the follow-up: turn inventory into interaction.

End-state diagram

sequenceDiagram
  participant Agent as Any agent
  participant MC as mission-control.db
  participant HM as hivemind.db
  participant Subs as subscriptions
  participant Auto as hive-handlers/*.sh
  participant NewMC as auto-created MC task
  participant LO as learning-opportunities/

  Agent->>MC: mc.js done <id>
  MC->>HM: post learning (T7)
  MC->>HM: post task-completion (T2)
  MC->>HM: post failed-task (T12, if outcome/reason matches failure regex)

  HM-->>Subs: SELECT WHERE kind=... AND enabled=1
  Subs-->>Auto: spawn callback (fire-and-forget, non-blocking)

  Auto->>NewMC: mc.js add with dedup (proveo QA / skillforge BookStack / codecraft bug)
  Auto->>LO: write lesson draft (for failed-task)

  Note over NewMC: original mc.js done returned long ago
  Note over LO: human reviews drafts

What changed

Phase 1 — Event bus live (the spine)

  • T1 — hivemind.js subscribe engine (CodeCraft, MC #8054): schema migrated with agent, kind, callback, enabled, correlation_filter. post now fires subscribers async + isolated; one callback failing doesn't stop the rest. subscribe, unsubscribe, subscriptions, fire subcommands.
  • T2 — 3 seed subscriptions (MC #8055): Proveo on task-completion, Skillforge on architecture-change, CodeCraft on error. Plus mc.js done now fires a task-completion intel alongside the pre-existing learning writeback.
  • T3 — auto-create MC tasks (MC #8056): the 3 subs now point to handler scripts at ~/system/tools/hive-handlers/*.sh which create properly owned MC tasks (QA review: #<id>, Update BookStack: <blueprint>, Investigate error intel#<id>) with dedup so duplicate events don't flood the queue. Audit log: ~/system/logs/hive-auto-route.log.

Phase 2 — Library activation

  • T4 — library auto-push daemon (FlowForge, MC #8057): ~/Library/LaunchAgents/com.alai.library-sync.plist runs library.js sync --fix every 5 min. Pre-flight snapshot at ~/system/backups/library-pre-autopush-20260417-0041/. Last-sync age went from 26 days → minutes.
  • T5 — FORGE first sync (MC #8058): BLOCKED-by-env. FORGE (10.0.0.2) unreachable from ANVIL (different subnet, no route). Documented in ~/system/ops/forge-connectivity-debt.md; follow-up MC #8070 for Alem to check physical/network state.
  • T6 — MCP distribution (Petter, MC #8059): decision doc ~/system/rules/mcp-distribution.md (93 lines). CodeCraft, Proveo, FlowForge got targeted MCP overlays (previously only Finverge had MCP). MCP column in library company-status went 1/13 → 4/13. Securion escalation: email MCP carries plaintext ALAI account credentials — Parisa Tabriz audit recommended.

Phase 3 — Skill usage visibility

  • T7 — PostToolUse hook (CodeCraft, MC #8060): ~/.claude/hooks/skill-use-counter.sh registered in settings.json. Every Skill invocation increments skill-registry.use_count. Non-blocking, exit 0 always. Companion tool: ~/system/tools/skill-usage.js (--top, --all, --dead).
  • T8 — weekly audit report (MC #8061): ~/system/tools/skill-audit-report.sh writes markdown to ~/system/reports/skill-audit-<date>.md. Scheduled via com.alai.skill-audit.plist Monday 07:00. First run found 62 retirement candidates (bootstrap — expected on week 1).

Phase 4 — Discover

  • T9 — persistent inverted index (AgentForge, MC #8062): ~/system/tools/.alai/discover-index.json (2.9MB, 521 entries across 6 sources). discover.js --rebuild-index flag. Post-sync wrapper ~/system/tools/library-sync-wrapper.sh runs library sync AND index rebuild. Query speed 200–500ms → <50ms. "drop" coverage 0 → 4 categories.
  • T10 — LightRAG fallback (MC #8063): when local hits < 3 AND !--no-lightrag, escalate to LightRAG semantic query with 5s timeout. Silent skip if LightRAG slow/unavailable. Results labelled LIGHTRAG (fallback — semantic).

Phase 5 — Meta-agent activation

  • T11 — daily meta-agent loop (AgentForge, MC #8064): ~/system/tools/meta-agent-loop.js runs daily 03:30 CEST via com.alai.meta-agent-loop.plist. Scans hivemind intel (learning + failed-task, last 24h), extracts bigrams, flags themes seen ≥ 3 times, creates NEW SKILL PROPOSAL MC tasks owned by skill-creator. Dedup across days by MC list grep. Never auto-commits — Alem approves.
  • T12 — failed-task auto-trigger (MC #8065): mc.js done with outcome/reason matching fail|error|broken|regression|workaround|bypass|skip|override|fix-later now posts a failed-task intel. learning-agent subscription invokes ~/system/tools/learning-opportunity-draft.sh which writes a lesson draft to ~/system/learning-opportunities/<task>-<ts>.md.

Active subscriptions (after Phase 1)

Agent Kind Handler
proveo task-completion hive-handlers/proveo-auto-qa.sh → QA review: #<id>
skillforge architecture-change hive-handlers/skillforge-auto-doc.sh → Update BookStack: <bp>
codecraft error hive-handlers/codecraft-auto-bug.sh → Investigate error intel#<id>
learning-agent learning (filter: FAILED) learning-opportunity-draft.sh → markdown draft

New daemons (launchd)

Label Schedule Script
com.alai.library-sync every 5 min library-sync-wrapper.sh (library sync + discover rebuild)
com.alai.skill-audit Monday 07:00 skill-audit-report.sh
com.alai.meta-agent-loop daily 03:30 meta-agent-loop.js

New knobs + files

  • ~/system/tools/.alai/discover-index.json — persistent search index (atomic writes)
  • ~/system/logs/hive-auto-route.log — audit log for every auto-created MC task
  • ~/system/logs/skill-use.log — every skill invocation timestamped
  • ~/system/logs/meta-agent-loop.log — daily meta-agent run output
  • ~/system/logs/learning-opportunity.log — failed-task → lesson draft audit
  • ~/system/learning-opportunities/ — lesson drafts (git-ignored)
  • ~/system/rules/mcp-distribution.md — MCP decision table (per-company)
  • ~/system/ops/forge-connectivity-debt.md — FORGE blocker write-up

Known issues (deliberate, not silent)

  • T5 FORGE sync blocked — MC #8070 open. No code fix; needs Alem to check FORGE network.
  • LightRAG probe still times out under heavy ingest (65K pending drain running). Fallback works but docker inspect intermittently shows unhealthy. Same root cause as System Evolution T1 — raise timeout or swap probe when convenient.
  • B2 offsite backup still paused (pre-existing quota). Not related to Hive but keeps the regression suite at 1 FAIL until resolved.
  • email MCP plaintext credentials — Petter flagged during T6. Recommend Securion audit before further rollout.
  • ZAKON PLAN drift in 8 legacy specs — linter reports WARN, not FAIL. Retroactive sweep low priority.
  • runbooks/hivemind-subscriptions.md — subscribe/unsubscribe/fire semantics, isolation model
  • runbooks/library-auto-push.md — daemon schedule, drift handling, snapshot rollback
  • runbooks/skill-use-counter.md — hook + skill-usage.js queries
  • runbooks/meta-agent-loop.md — bigram approach, dedup rules, Alem-approval gate
  • runbooks/discover-reindex.md — index structure, rebuild flow, LightRAG fallback
Back to top