Imported from weiyangzen/b3ehive (SKILL.md). Install upstream with npx skills add weiyangzen/b3ehive. Copyright stays with the author.

name: b3ehive description: Root guide for b3ehive's five portable swarm skills: compete, execution, learn, optimization, and looper cron builders for Codex, Claude Code, Cursor, Grok Build, opencode, OpenClaw, and Hermes.

b3ehive Skill Index

b3ehive provides five portable SKILL.md directories:

• compete-cron-builder: bounded proposal competitions with n workers, m proposals, choose k, all-valid coverage union, repair queues, blueprint synthesis, and three-way challenge artifact coverage.
• execution-cron-builder: blueprint-driven DAG execution with worker/master cursors, validation gates, checkpointing, and cleanup.
• learn-cron-builder: source-to-target learning cron for code understanding, subset learning, code-to-code transform, and human-language translation.
• optimization-cron-builder: design-guided architecture and optimization cron.
• looper-cron-builder: resource-aware bridge controller for DAG nodes, BridgeSurfaces, BridgeMetrics, ResourceEnvelopes, ResourceLeases, SideEffectGates, OperatorSignals, nested run ledgers, compact evidence, reward ledgers, ROI, and no-reward pause.

Shared invariants across all five skills:

• [ ], [_], and [x] are the only authoritative acceptance states.
• Runtime labels such as paused, cancelled, done, accepted, or finalized are telemetry unless paired with the checkbox state.
• Auto choices should use EstimatorPolicy; hardcoded values must be hard caps for AI boundaries, safety, compatibility, or operator overrides.
• Nontrivial skill selection, model/provider choice, and nested skill composition should leave a RouteDecision.
• Nested skill calls require route, lease, evidence, side-effect, depth, and ROI controls; child skills cannot write [x].
• Evidence lint, not narrative confidence, gates acceptance.
• LooperLog is the shared multi-grain feedback surface for TargetObject movement and InstrumentObject quality across micro, skill, composition, scaffold, tool, and task grains. It is evidence/backlog, not accepted policy change.
• Normal execution may change the target object inside the task boundary. Instrument changes to skills, scaffolds, validators, routes, ledgers, scripts, or tools require a separate [ ] -> [_] -> [x] improvement lifecycle.
• Self-evolution requires EvidenceLint, ROI, ParetoGate, rollback, and master [x].
• Prompt blocks, hooks, validators, and coding-tool glue are generated engineering artifacts unless promoted by evidence; do not grow a prompt/hook forest inside the five core skills.

Read looper-cron-builder/references/b3ehive-bridge-contract.md for the shared contract when updating skills, generating coding-tool adapters, using nested skills, or designing route/evidence/looper_log/ROI/self-evolution behavior.