swarm

name	swarm
description	Speculative swarm — forks N agents to explore divergent strategies for the same task, cross-pollinates insights at checkpoints, prunes convergent branches, and fuses the best fragments into a composite result. Use when a problem has multiple viable approaches and you want to explore them simultaneously, or when the user invokes /swarm run <spec-path>.

Speculative Swarm Skill

Parallel FORK → CHECKPOINT → PRUNE → MERGE pipeline for divergent exploration.

Usage

/swarm run <spec-path>

Example:

/swarm run specs/044-factory-skill-mvp/README.md

Execution

This skill is orchestrated by the pipeline engine. When invoked, execute:

synodic harness run --pipeline swarm --spec <spec-path>

Pipeline definition: .harness/pipelines/swarm.yml Prompt templates: skills/swarm/prompts/ Output schemas: schemas/strategy-set.json, schemas/branch-report.json, schemas/merge-report.json

Pipeline Steps

1. STRATEGIZE — Generate N divergent strategies (algorithm, architecture, data structure, paradigm)
2. EXPLORE — Fan-out branch solving in isolated worktrees (parallel)
3. CHECKPOINT — Jaccard similarity on changed file sets; cross-pollinate insights
4. PRUNE — Kill convergent branches (similarity > threshold); min 2 survivors
5. MERGE — Fragment-fusion of best parts from surviving branches
6. MERGE GATE — Preflight checks on merged result
7. CREATE PR — Push and create pull request

Algorithmic Spine

Deterministic CLI commands (zero LLM cost):

• synodic swarm checkpoint — Pairwise Jaccard similarity + cross-pollination
• synodic swarm prune — Convergence detection with min 2 survivors

Governance

• Checkpoint map: CHECKPOINT → Layer 1, MERGE GATE → Layer 2
• Governance log: .harness/swarm.governance.jsonl
• See HARNESS.md for the governance protocol