fractal

name	fractal
description	Fractal decomposition — recursively split a complex task into orthogonal sub-specs, solve each leaf independently via subagents, then reunify results bottom-up. Use when a task is too large for a single agent pass, when you need to decompose a problem into independently-solvable pieces, or when the user invokes /fractal decompose <task-or-spec-path>.

Fractal Decomposition Skill

Recursive DECOMPOSE → SOLVE → REUNIFY tree for complex tasks that are too large for a single agent pass.

Usage

/fractal decompose <task-or-spec-path>

Example:

/fractal decompose specs/058-code-harness-orchestration/README.md

Execution

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

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

Pipeline definition: .harness/pipelines/fractal.yml Prompt templates: skills/fractal/prompts/ Output schemas: schemas/decompose-verdict.json, schemas/solve-report.json

Pipeline Steps

1. COMPLEXITY CHECK — Deterministic scoring; auto-LEAF if below threshold
2. DECOMPOSE — Recursive split into orthogonal sub-tasks (fan/loop)
3. DECOMPOSE GATE — TF-IDF orthogonality, cycle detection, budget allocation
4. SCHEDULE — DAG topological sort into parallel execution waves
5. SOLVE — Fan-out leaf solving in isolated worktrees
6. SOLVE GATE — Preflight checks per leaf
7. REUNIFY — Algorithmic git merge + AI conflict resolution
8. PRUNE — Set cover redundancy detection
9. CREATE PR — Push and create pull request

Algorithmic Spine

Deterministic CLI commands (zero LLM cost):

• synodic fractal complexity — Score spec complexity
• synodic fractal gate — TF-IDF orthogonality + cycle detection
• synodic fractal schedule — Topological sort → waves
• synodic fractal reunify — git merge-tree conflict detection
• synodic fractal prune — Set cover redundancy

Governance

• Checkpoint map: DECOMPOSE GATE → Layer 1, SOLVE GATE → Layer 1/2, REUNIFY → Layer 2
• Governance log: .harness/fractal.governance.jsonl
• See HARNESS.md for the governance protocol