Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
What pattern keeps being produced?
What structure produces it?
What feedback loops stabilize or amplify it?
Where are the delays, incentives, stocks, flows, and information gaps?
What intervention changes the system without making it worse?
Core doctrine
A system is connected parts producing a pattern.
Use $cybernetic when repeated events imply structure.
Do not use $cybernetic for isolated local work.
Do not optimize a part while damaging the whole.
Do not mistake an event for the system.
Do not use a clear-system checklist in a complex system.
Do not use analysis paralysis in chaos.
Do not intervene before identifying feedback, delay, incentives, and leverage level.
Integration contract
$cybernetic classifies the system and selects leverage. It does not implement.
If used after a same-cluster stop rule, local_patch_allowed should usually be no unless the context proves the selected owner mutation is a system-level normal form rather than another local point fix.
Cause and effect are obvious, stable, and directly observable.
Use:
checklist
standard work
precision
known process
Avoid overthinking and unnecessary innovation.
Complicated
Cause and effect exist but require analysis, decomposition, or expertise.
Use:
diagnosis
specialist expertise
modeling
root-cause analysis
comparison of alternatives
Avoid guessing and generic experts.
Complex
Cause and effect are emergent and visible mostly in hindsight.
Use:
safe-to-fail experiments
small probes
monitoring
adaptation
diverse perspectives
directional strategy
Avoid rigid plans, single-point forecasts, and pretending to control the system.
Chaotic
Cause and effect are broken, hidden, or changing too fast to reason through.
Use:
stabilize first
create safety
act quickly
bound harm
then sense and analyze
Avoid analysis paralysis.
Mixed
Many real situations contain different subsystem types. Split them.
Example:
technical rollout: complicated
human adoption: complex
incident response: chaotic until stabilized
checklist procedure: clear
DART diagnostic
Use DART to quickly orient:
D — Deconstruct: What are the parts, stocks, flows, and boundaries?
A — Analyze: What is the cause/effect relationship type?
R — Recognize: What pattern, archetype, or prior system does this resemble?
T — Test: What is the smallest safe probe or stabilizing action?
In chaos, T means stabilize first rather than experiment.
Meadows ladder
Prefer higher leverage when available, but stay humble.
Approximate leverage order, from lower to higher:
parameters / numbers
buffers
stock-flow structure
delays
balancing feedback
reinforcing feedback
information flows
rules / incentives / constraints
power to self-organize
goals
paradigms / mental models
ability to transcend paradigms
Rules:
Do not obsess over parameters when structure, rules, goals, or mental models are producing the pattern.
Do not jump to paradigm change when a missing information flow would solve the issue.
Do not change goals or rules without scanning incentives, power, and second-order effects.
Do not ignore delays; delayed feedback can make good actions look bad and bad actions look good.
Iceberg model
When the user gives an event, move down the iceberg:
Event: What just happened?
Pattern: What keeps happening over time?
Structure: What rules, incentives, flows, delays, and power relations produce it?
Mental model: What beliefs, goals, identities, or paradigms make the structure seem natural?
Do not stop at event-level explanation unless the system is clear and local.
Stocks and flows
Ask:
What is accumulating?
What drains it?
What fills it?
What is the buffer?
What is the bottleneck?
What is the delay between action and observable effect?
What stock is being ignored because it is intangible?
feedback routed to someone without decision rights;
proxy metric replacing goal;
response too strong or too weak;
balancing loop overwhelmed by reinforcing loop.
Incentive and cobra-effect scan
Before recommending a metric, reward, target, policy, or automation, ask:
What behavior will this reward?
What proxy replaces the real goal?
Who can game it?
What gets worse if people optimize for the measure?
What delayed harm might appear after the success metric improves?
If the proxy can beat the purpose, redesign the rule or measure.
Platform view
When inside the system, seek outside perspective through:
mentor / outsider / stakeholder not captured by the current incentive
data / measurement independent of narrative
time / before-after and trend comparison
If all evidence comes from inside the current mental model, mark confidence lower.
Intervention design
Match intervention to system type:
| System type | Best first move | Bad first move |
|---|---|
| clear | checklist / standard process | clever reinvention |
| complicated | analysis / expertise | generic intuition |
| complex | safe-to-fail probes / adaptation | rigid master plan |
| chaotic | stabilize / create safety | analysis paralysis |
Output modes
Standard
Use sections:
System boundary
System type
Events -> patterns -> structure -> mental models
Stocks / flows / feedback
Incentives and delays
Leverage map
Intervention
Probe / monitoring plan
Cybernetic Bottom Line
Fast
Use:
System type:
Pattern:
Leverage:
Move:
Watch:
Cybernetic Bottom Line:
