An agent skill that guides users through systems thinking using the framework from Donella Meadows' Thinking in Systems: A Primer.

The skill helps users move beyond surface-level events to see the feedback loops, stocks, flows, and mental models that produce the patterns they observe. It follows a Socratic approach — asking questions before asserting, letting the user build understanding rather than being told.

When a user describes a situation — why something keeps happening, why a fix didn't work, how a system behaves — this skill guides them through a structured exploration:

1. Is it a system? — Identify elements, interconnections, and purpose
2. Stocks and flows — Map what accumulates and what changes
3. Test the model — Challenge the model against reality
4. Why systems work well — Understand resilience, self-organization, hierarchy
5. Why systems surprise us — Recognize boundaries, delays, bounded rationality
6. System traps — Identify common structural traps and their opportunities
7. Leverage points — Find where small shifts produce big changes

The skill produces structural diagrams (stocks, flows, feedback loops) and behavior-over-time graphs throughout the exploration.

npx skills add chenhunghan/thinking-in-systems-skill

The skill activates when a user:

• Describes observations about how something behaves
• Asks why something keeps happening
• Wants to understand the structure behind patterns
• Wants to model a system
• Describes recurring patterns or unintended consequences
• Describes situations where fixes don't seem to work
• Focuses too much on events and needs help seeing deeper structure

thinking-in-systems-skill/        # repo root
  README.md
  LICENSE
  .gitignore
  run-evals.sh                    # eval runner (run from repo root)
  thinking-in-systems/            # skill directory
    SKILL.md                      # Skill definition (frontmatter + instructions)
    phases/                       # Phase reference files read during exploration
      01_is_it_a_system.md
      02_stocks_and_flows.md
      03_test_the_model.md
      04_why_systems_work_well.md
      05_why_systems_surprise_us.md
      06_system_traps.md
      07_leverage_points.md
    evals/                        # Eval definitions
      evals.json                  # 10 task evals, 75 expectations
      trigger-evals.json          # 28 trigger boundary tests

Prerequisites: claude CLI installed and authenticated.

# from repo root
chmod +x run-evals.sh
./run-evals.sh

Results are written to thinking-in-systems-workspace/eval-run-N/ (gitignored). Each eval produces a response file and a grade file with pass/fail per expectation.

MIT