systems-thinking - SKILL.md Agent Skill

license: Apache-2.0 name: systems-thinking description: Analyze complex systems through stocks, flows, and feedback loops to find high-leverage interventions. For organizational, environmental, social, and technical systems exhibiting circular causality. NOT for linear problems or simple cause-effect chains. allowed-tools: Read category: Cognitive Science & Decision Making tags: - systems-thinking - mental-models - complexity - feedback-loops - holistic

Systems Thinking

Diagnose why systems cause their own behavior and identify structural interventions that produce sustainable change.

Decision Points

Pattern Recognition Table

If you see this behavior...	Then check for...	Try this intervention tree
Oscillation (boom-bust cycles)	Excessive delays in corrective feedback	Reduce delays OR dampen response rates
Exponential growth hitting limits	Balancing loop activating	Expand limits OR reduce growth rate early
Multiple failed fixes	Policy resistance trap	Find shared overarching goal
Performance declining over time	Drift to low performance	Hold absolute standards vs. relative
Resource degradation	Tragedy of commons	Create direct user feedback

Intervention Leverage Decision Tree

Problem identified → Map stock-flow structure first
│
├─ High leverage available?
│  ├─ Can change paradigm/worldview? → Transform mental models
│  ├─ Can shift system goals? → Redefine success metrics  
│  ├─ Can add/strengthen feedback loops? → Create information flow
│  └─ Can change rules? → Restructure incentives
│
├─ Medium leverage only?
│  ├─ Can improve information flow? → Connect decision-makers to consequences
│  └─ Can adjust parameters? → Change numbers/rates (lowest leverage)
│
└─ No structural leverage?
   └─ Wrong problem OR linear system → Use different approach

Trap Escape Decision Matrix

Trap Type → First Check → If Yes → If No
Policy Resistance → Others resisting your solution? → Find shared goal → Push harder (escalates)
Tragedy of Commons → Shared resource degrading? → Create ownership/feedback → Regulate only
Addiction → Intervention creating dependency? → Strengthen original capacity first → Continue intervention
Escalation → Competition intensifying? → Unilateral restraint OR negotiation → Try to win (unsustainable)

Failure Modes

Event Fixation

Detection: You're analyzing who did what when, looking for someone to blame
Symptom: "If only we fire X/hire Y/change Z, the problem will be solved"
Fix: Draw behavior over time graphs; map the structure generating events

Parameter Tweaking

Detection: 90% of discussion focuses on adjusting numbers (budgets, rates, standards)
Symptom: "We need to increase/decrease the target by X%"
Fix: Ask "What structure is producing these numbers?" Map information flows and feedback loops

Linear Causality Trap

Detection: Expecting proportional responses; surprised by sudden behavioral shifts
Symptom: "We did X, so Y should happen proportionally"
Fix: Map circular causality; identify reinforcing loops that create exponential effects

Control Obsession

Detection: Demanding predictable outcomes; treating uncertainty as failure
Symptom: "We need better forecasting/control systems"
Fix: Design adaptive feedback policies instead of rigid controls

Symptom Relief Addiction

Detection: Quick fixes that need repeating; original problem capacity atrophying
Symptom: "The intervention is working, we just need to do more of it"
Fix: Strengthen the system's original capacity; plan intervention withdrawal

Worked Examples

Example 1: Organizational Overtime Crisis

Situation: Software team chronically missing deadlines despite working 60+ hour weeks

Novice approach: Hire more developers, mandate better time estimation Systems analysis:

Stock: Work backlog accumulating faster than completion rate
Inflows: New features, bug reports, scope changes
Outflows: Completed work (declining due to exhaustion/turnover)
Feedback loop: Overtime → fatigue → more bugs → more rework → more overtime

Intervention chosen: Slow intake rate (say no to new requests) + improve quality to reduce rework Outcome: Backlog initially grew (counterintuitive) but outflow rate increased as bugs decreased Unintended consequence: Sales team frustrated by delayed features, required stakeholder alignment

Example 2: Environmental Resource Depletion

Situation: Fishing community experiencing declining catch despite harder work

Stock-flow mapping:

Stock: Fish population (declining)
Inflows: Fish reproduction (slow, 2-year delay)
Outflows: Fishing harvest (increasing with effort)
Structure: Classic tragedy of commons - individual rationality, collective irrationality

Decision point navigation:

Oscillation pattern? → Check delays: Yes, 2-year reproduction lag
Shared resource? → Yes, tragedy of commons trap
High leverage intervention? → Create direct feedback between individual action and consequences

Intervention chosen: Assign fishing territories (privatization) + seasonal quotas based on stock levels Outcome: Short-term income drop, long-term sustainability
Unintended consequence: Some fishers excluded from system, required compensation mechanism

Example 3: Technology Performance Degradation

Situation: Database system slowing down despite hardware upgrades

Systems lens applied:

Behavior over time: Response time increasing exponentially under load
Stock: Query complexity accumulating in system
Structure: Performance fixes create more complex queries → slower performance → more "optimization"

Leverage points tested:

Parameter level: More RAM/CPU → Temporary improvement only
Information flow: Real-time performance visibility to developers → Better query design
Rules change: Query complexity limits → Structural improvement

Result: Information flow change had highest leverage - developers changed behavior when they saw real impact

Quality Gates

Behavior over time graphs generated for key variables (not just snapshots)
All major stocks identified and quantified with actual numbers where possible
All inflows and outflows mapped with approximate rates
At least 2 feedback loops identified (1 reinforcing, 1 balancing minimum)
Delays between actions and consequences measured/estimated
System archetype/trap pattern recognized and named
At least 2 different leverage points tested (not just parameter changes)
Unintended consequences anticipated and mitigation planned
Success metrics aligned with actual system purpose (not just activity measures)
Intervention includes feedback mechanism for course correction

NOT-FOR Boundaries

Don't use systems thinking for:

Linear technical problems: Use root cause analysis instead
One-time events: Use project management for discrete deliverables
Emergency response: Use crisis management protocols for immediate threats
Simple optimization: Use operations research for well-defined mathematical optimization
Individual behavior change: Use psychology/coaching for personal development

Delegate to other skills:

Data analysis: For statistical correlation without feedback loops
Strategic planning: For competitive positioning and market analysis
Process improvement: For workflow optimization without complex interdependencies
Negotiation: For resolving conflicts between known parties with clear interests

Systems thinking boundaries:

Requires circular causality and feedback loops
Needs time delays between cause and effect
Involves multiple stakeholders with competing goals
Exhibits unintended consequences from well-intentioned actions
Shows persistent problems despite repeated interventions