← ClaudeAtlas

task-relationship-analysislisted

Use BEFORE executing any multi-step or multi-feature request (3+ tasks). Forces a pre-execution pass that maps how the tasks relate — synergies, conflicts, shared substrate, ordering — so you don't mechanically execute them in listed order and miss that several should be built on one shared piece, or that two conflict. Scaffolds a pairwise matrix + synthesis checklist.
jajupmochi/agent-harness · ★ 1 · AI & Automation · score 75
Install: claude install-skill jajupmochi/agent-harness
# task-relationship-analysis Overhaul task 15. Some agents/models execute a numbered task list **mechanically, in order**, and miss the cross-task picture — e.g. three separate "add a sub-agent" asks that should be **one** framework, or two tasks that **conflict** if done independently. This skill makes the relationship analysis a required first step. ## When Any request with **3+ tasks**, or multiple logically separable features, before you start executing. ## Do this first 1. List the tasks, then scaffold the analysis: ``` python3 scripts/scaffold.py "task A" "task B" "task C" … # or pipe one per line on stdin ``` It emits every **unordered pair** exactly once (nothing skipped) + a synthesis checklist. 2. Fill each pair with one relationship: **SYN** (synergy / share substrate) · **CONF** (conflict) · **SEQ** (ordering dependency) · **INDEP** (independent). 3. Answer the checklist: shared substrate to unify · conflicts to resolve in design · mergeable work · ordering/DAG from the SEQ pairs · cross-cutting constraints to do first. 4. Only THEN execute — in the order the DAG implies, unifying what the matrix flagged. ## Why deterministic scaffold + LLM analysis `scaffold.py` is pure code: it guarantees the shape and that **every pair is considered**. The model does the actual judgement (which pair is SYN vs CONF). LLM-as-component, not the core. ## Example payoff (this repo's own overhaul) The 16-point overhaul's matrix surfaced that tasks 2/3/