/SystemsThinking

name: SystemsThinking
evals:
  manual:
    reason: "feedback structures and leverage points must be judged against the supplied system evidence and behavior"
    falsifier: "the reviewer finds a causal link unsupported, a loop direction wrong, or a proposed leverage point cannot affect the stated outcome"
metadata:
  short-description: "Analyze system structure, feedback loops, archetypes, and leverage points behind recurring behavior. Use RootCauseAnalysis for incident causal chains."
version: 1.1.1
description: "Structural analysis of complex systems — Iceberg model, Causal Loop feedback diagrams, archetype matching, Meadows leverage points, and concept maps — grounded in the premise that behavior is generated by structure. USE WHEN systems thinking, causal loop, feedback loops, archetypes, leverage points, iceberg model, fix the system, why does this keep happening, recurring problem, second-order effects. NOT FOR incident causal chains (use RootCauseAnalysis)."

SystemsThinking Skill

SystemsThinking explains recurring behavior through the structure that generates it: iceberg layers, causal loops, archetypes, leverage points and concept maps. Most real runs point it at a recurring problem in the principal's own harness (a metric drifting the wrong way, a library or hook set that keeps growing, a signal that died without anyone noticing), usually chaining CausalLoop, FindArchetype and FindLeverage, with the measured facts already in the request; a few take a real-world system such as a city's water supply or a team's on-call rotation. Many ask for a brief inline pass whose output feeds a larger report. During an acute incident, stabilize first and analyze afterward; incident causal chains belong to RootCauseAnalysis. Art renders the diagrams, and BeCreative generates candidates at a chosen leverage level in one pass, and HillClimb Ideate evolves them over several rounds.

Ideal State

• Every analysis explains the behavior through the structure that generates it (loops, delays, incentives, flows, beliefs), never through the last event or a person.
• The analysis starts from the facts and paths the request supplies and reads those paths rather than searching for them again. A supplied fact the conclusion rests on is checked at its source, a caller's hypothesis is never taken as fact, and a missing loop link or measured fact is read or measured at its source rather than inferred; any link still inferred is marked. Any search names the paths it covers, never the whole home tree or ~/Projects.
• Evidence and inferred mechanism are told apart in every output, and every recommended intervention is traced through at least one round of its own feedback, with its second-order effects named, before it ships.
• A recommendation is the highest-leverage change that is actually feasible, with a concrete first move and a date to re-evaluate it.
• The recommendation names the specific file, hook, rule, metric or mechanism to change, and carries the input's measured facts (counts, rates, dates) through to it, so its effect can later be checked against the same numbers.
• The actors are whatever actually acts in the system. In a solo operator's AI harness that is hooks, models, doctrine files, scheduled jobs and the principal; an organization, politics or culture is never invented to fit the frame.

Intent Routing

| Intent | Trigger | File | |----------|---------|------| | **Iceberg** | "iceberg model", "structural cause", "why does this keep happening", walk from symptom down to structure | `Intents/Iceberg.md` | | **CausalLoop** | "causal loop", "feedback loop", "connection circle", "map relationships", build a CLD | `Intents/CausalLoop.md` | | **FindArchetype** | "systems archetype", "recognize this pattern", "fixes that fail", "shifting the burden", "tragedy of the commons" | `Intents/FindArchetype.md` | | **FindLeverage** | "leverage point", "where to intervene", "highest-leverage change", Meadows' 12 | `Intents/FindLeverage.md` | | **ConceptMap** | "concept map", "map the entities", "relationship map", Novak-style mapping | `Intents/ConceptMap.md` |

| File | Read when | |------|-----------| | `Foundation.md` | boundaries are contested, structures are informal, or the dynamics may mislead (Iceberg, CausalLoop) | | `Archetypes.md` | matching an archetype: discriminators and intervention safeguards (FindArchetype) | | `LeveragePoints.md` | ranking leverage: exceptions and intervention gates (FindLeverage) |

Data & Config

User customizations, when present, live at `<configRoot>/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/SystemsThinking/` (configRoot is `$CLAUDE_CONFIG_DIR` when set, else `~/.claude`). Any `PREFERENCES.md`, configurations or resources there override the defaults; when the directory does not exist, proceed with skill defaults. A `History.md` there, when present, records the install's own past runs and their measured facts; nothing reads it at run time.

Gotchas

_No gotchas documented yet. Add failures here as they're discovered._

Examples

User: "Analyze the model router's feedback structure after a model swap: the router issues binding routes, the new main model declines about X% of them inline, and the delegation rate fell from A% to B%."
→ CausalLoop, then FindLeverage
→ Named R/B loops that keep work on the main model, and one leverage point naming the rule or meter to change, judged against the A%→B% figure

User: "Why does the LifeOS skill library keep accumulating overlapping skills (N now), and what structural leverage point would stop the sprawl rather than just pruning it once?"
→ Iceberg, FindArchetype, FindLeverage
→ Events to pattern to the structure that adds skills, the archetype it matches, and a structural change with a first move and a re-check date

User: "Why does a growing city keep running short of water after every reservoir expansion: feedback structure of supply, demand, pricing and drought"
→ CausalLoop, FindArchetype,