HivemindOS manual

Hive Skills

Hive skills are HivemindOS-native procedures. They teach agents how to use the hive’s own memory, tools, apps, runtimes, projects, and workflows.

These skills are auto-installed into the shared brain because they are foundational.

Skill Purpose
engineering-discipline Selects a proportionate engineering path and anchors Work Board evidence: design only when needed, real baselines, red/green or non-applicability receipts, focused verification, independent review, rollback, and repository state.
bankr-skill-deployment Chooses the correct Bankr skill, app, x402, Wallet/Agent API, and always-on hosting boundaries, then packages, publishes, installs, updates, and verifies the capability end to end.
create-zero-human-company Creates a durable company from a business goal or existing repository, adds an available crew, measurable apex goal, products, budgets, and approval policies, verifies the saved record, and leaves autonomy stopped until the operator explicitly launches it.
operate-zero-human-company Operates an already-created company through a strict highest-leverage triage order — setup blockers, expiring spend approvals, genuine ACTION NEEDED decisions, retryable infrastructure noise, budget and burn review, directive hygiene, and launch/pause/freeze posture — without bypassing approval gates or clearing queues dishonestly.
hive-assimilate Mandatory pre-build search across shared brain, user projects, local/private indexes, and public GitHub before software creation.
hive-pulse Built-in last-30-days signal briefs across Reddit, X, YouTube, TikTok, Hacker News, Polymarket, GitHub, and web sources, powered by a pinned MIT licensed last30days engine.
hive-quant-research Runs research-only typed strategy families through lagged Rust simulation, independent Python statistics, overfitting controls, regime and factor audits, durable lineage, and approved scheduled runs.
hive-capability-search Discovers available tools, shared skills, optional packaged workflow playbooks, apps, agents, credentials by key name, and delivery channels such as slash commands, API routes, MCP tools, CLIs, and dashboard surfaces for a task.
hive-remote-capability-use Executes remote connected apps and fleet capabilities selected by capability search, including fresh discovery, Hivemind Link app-proxy routing, private file transfer, artifact verification, and side-effect gates.
hive-skill-fusion Turns a capability request into a reusable shared-brain skill.
hive-workflow-fusion Composes multi-step hive workflows from skills, apps, agents, and tools.
hive-aeon-fusion Converts reusable hive workflows into AEON-ready agent duties when appropriate.
hive-brain-memory Teaches agents how to recall canonical typed memory, keep operational receipts separate, review mined patterns, and evolve stale memories while preserving superseded history.
hive-brain-compiled-wiki Teaches agents how to compile durable findings into HivemindOS entity/concept/summary wiki pages, search/query the compiled graph through MCP, repair wiki health, and respect human collective shared-brain mirrors without restricting normal agent-to-agent work.
hive-skill-autoresearch Improves installed skills through a measured baseline, four independent variants, regression floors, independent review, and a human-approved winning diff. It works with native HivemindOS agents and can use Evo when a repository benchmark is available.
wrapup On an explicit session wrap-up, saves only durable reviewed facts through typed Shared Brain Memory and adds a concise redacted session summary to the user’s verified NotebookLM AI Brain notebook.

Supporting Hive Search Commands

There is no separate packaged skill named hive-find or hive-search in this repo. The canonical shared-brain memory search surface is the installed hive-brain CLI:

hive-brain answer "query"
hive-brain recall "query" --scope full-vault --limit 8
hive-brain evolve --memory-id mem-... --content "Updated durable memory"

hive-brain answer returns a concise grounded answer. hive-brain recall returns a hit list. Both try the running HivemindOS brain API first and fall back to local vault/index search, so raw agents can use the same private brain without needing dashboard-routed context. hive-brain record-operation keeps routine receipts in a bounded local journal, while hive-brain mine-patterns dry-runs recurring-learning, skill, and routine proposals before an explicit Brain Review enqueue. Durable writes use canonical memory heads, and hive-brain evolve replaces reviewed stale memory while preserving superseded history. Use the auto-installed hive-brain-memory skill for the full policy.

For durable synthesized wiki knowledge, use the auto-installed hive-brain-compiled-wiki skill instead of broad recall. It teaches agents to call brain_search_knowledge first for compiled entity, concept, and summary pages, then follow up with brain_get_node, brain_get_backlinks, or brain_graph_overview when they need graph detail.

For capability routing, setup also installs:

hive-capability-search "task or goal"
hive-capability-search --json --kinds skill,tool-schema "task or goal"
hive-capability-search --live "task or goal"

The command searches the local Context Index without needing a running dashboard. --live asks the authenticated dashboard /api/context-index route first so connected-app endpoint freshness can be included, then falls back to the local index unless --require-api is set.

Agents without shell access, including app-routed or phone-hosted agents, should not treat a missing CLI as a failed skill. They should use already-injected capability-search context, an authenticated /api/context-index bridge, or an exposed MCP/context-index tool. If none are available, they should build the map from visible skills/tools and record the retrieval gap.

Hive Assimilate

hive-assimilate is the default build gate for software work.

Agents using it should search in this order:

  1. User-pinned sources, such as a repo, path, PR, note, or shared skill.
  2. Shared brain recall with hive-brain answer "<query>" --scope full-vault.
  3. Relevant current workspace files, docs, tests, and conventions.
  4. User project roots when relevant and bounded.
  5. Local/private assimilation indexes.
  6. Public GitHub candidates, star-sorted for broad discovery.

The skill writes ASSIMILATION_LOG.md, ASSIMILATION_LOG.jsonl, and optional ASSIMILATION.json in the target project so agents can review what was searched, reused, or rejected.

Existing local candidate caches remain valid so project references do not break.

Hive Pulse

hive-pulse is the default Hive research pulse for current public signal.

It ships in packaged-skills/auto-install/hive-pulse/ with the pinned MIT licensed mvanhorn/last30days-skill engine bundled inside the skill. Fresh setup copies it into the shared brain, so agents do not need a separate upstream install before running a pulse brief.

Setup also installs the hive-pulse command shim. The shim loads shared Hive env keys, defaults browser-cookie reads off, uses the zero-config source set, supplies a deterministic query plan, and selects local deterministic judging unless the user explicitly configures an LLM provider.

The default posture is read-only research. Reddit public paths, Hacker News, Polymarket public odds, GitHub public search, and available web paths can work without extra setup. User-owned keys or browser sessions expand coverage for X, YouTube, TikTok, Instagram, Threads, Pinterest, and richer web providers. Agents must check credentials by key name only and must not print secret values.

Source provenance and audit notes live in:

packaged-skills/auto-install/hive-pulse/.hivemind-skill-source.json

Bankr Skill Deployment

bankr-skill-deployment is the default delivery playbook when a capability should run with Bankr. It distinguishes a per-wallet skill, a request-time paid x402 endpoint, a wallet-aware Bankr app, and a separately hosted always-on backend before implementation begins.

The skill includes package/version verification, public-GitHub and Bankr Agent API installation gates, names-only secret handling, direct Wallet API versus asynchronous Agent API guidance, and real-path x402 schema/payment verification. It also captures the managed copy trader’s durable integration rules: existing Shared Hive Env selections continue without being re-saved, new variables use Save, company restrictions require a validated company task, and post-trade fees settle only after independent transaction verification.

Source Of Truth

The repository source is:

packaged-skills/auto-install/bankr-skill-deployment/SKILL.md
packaged-skills/auto-install/create-zero-human-company/SKILL.md
packaged-skills/auto-install/hive-assimilate/SKILL.md
packaged-skills/auto-install/hive-pulse/SKILL.md
packaged-skills/auto-install/hive-quant-research/SKILL.md
packaged-skills/auto-install/hive-capability-search/SKILL.md
packaged-skills/auto-install/hive-remote-capability-use/SKILL.md
packaged-skills/auto-install/hive-skill-fusion/SKILL.md
packaged-skills/auto-install/hive-workflow-fusion/SKILL.md
packaged-skills/auto-install/hive-aeon-fusion/SKILL.md
packaged-skills/auto-install/hive-brain-memory/SKILL.md
packaged-skills/auto-install/hive-brain-compiled-wiki/SKILL.md
packaged-skills/auto-install/hive-skill-autoresearch/SKILL.md
packaged-skills/auto-install/wrapup/SKILL.md

Setup mirrors it into the shared brain at:

Skills/bankr-skill-deployment/SKILL.md
Skills/create-zero-human-company/SKILL.md
Skills/hive-assimilate/SKILL.md
Skills/hive-pulse/SKILL.md
Skills/hive-quant-research/SKILL.md
Skills/hive-capability-search/SKILL.md
Skills/hive-remote-capability-use/SKILL.md
Skills/hive-skill-fusion/SKILL.md
Skills/hive-workflow-fusion/SKILL.md
Skills/hive-aeon-fusion/SKILL.md
Skills/hive-brain-memory/SKILL.md
Skills/hive-brain-compiled-wiki/SKILL.md
Skills/hive-skill-autoresearch/SKILL.md
Skills/wrapup/SKILL.md

Do not create duplicate shared-brain copies of the same Hive skill. If an imported runtime carries an equivalent workflow under another name, keep hive-assimilate as the canonical shared-brain skill.

Token Savings

Hive skills are also cost controls. See Token And Cost Savings for the full loop: hive-brain recall, hive-capability-search, hive-assimilate, karpathy-guidelines, Hive Fusion, and runtime usage analytics.

Expanded image Scroll to pan · Esc to close
100%