agentleFS
calvyntwhcalvyntwh/karu-custom-skills/skills/second-order-thinking/SKILL.md

second-order-thinking

Skill1 starsChanged 2 months agoMITView on GitHub

Consider cascading consequences of decisions across time horizons. Ask "And then what?" to anticipate second and third-order effects. Use when making important decisions, planning projects, or analyzing outcomes that seem counterintuitive. Distinct from inversion-thinking (which asks "How could this fail?") - second-order asks "What happens next, and next?

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

10 rules. 0 are common, 10 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 10

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Decision Making: When choosing between options with uncertain outcomes

    line 19
  2. Distinctive:

    Planning: When a decision will cascade through systems over time

    line 20
  3. Distinctive:

    Analysis: When an outcome seems counterintuitive (success or failure)

    line 21
  4. Distinctive:

    Strategy: When first-order wins lead to long-term losses (or vice versa)

    line 22
  5. Distinctive:

    Stop when effects become speculative or plateau.

    line 40
  6. Distinctive:

    Stop tracing chains when:

    line 81
  7. Distinctive:

    Before accepting any chain, ask:

    line 106
  8. Distinctive:

    Do actors in the chain have incentives that might break the link?

    line 108
  9. Distinctive:

    Combined Protocol: Run inversion to identify failure modes, then run second-order to trace what happens after those failures.

    line 123
  10. Distinctive:

    After a decision plays out, log only meaningful misses:

    line 137

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Second-Order Thinking

The ability to think beyond immediate consequences to cascading effects across time.

"First-level thinking is simplistic and superficial... Second-level thinking is deep, complex and convoluted." — Howard Marks, The Most Important Thing

"Failing to consider second- and third-order consequences is the cause of a lot of painfully bad decisions." — Ray Dalio

When to Use

  • Decision Making: When choosing between options with uncertain outcomes
  • Planning: When a decision will cascade through systems over time
  • Analysis: When an outcome seems counterintuitive (success or failure)
  • Strategy: When first-order wins lead to long-term losses (or vice versa)

When NOT to Use

  • Reversible, low-stakes decisions — first-order thinking suffices.
  • Time-sensitive decisions — analysis overhead exceeds decision value.
  • Well-underbounded problems — where cascade chains are infinite and unpredictable.
  • When already using inversion-thinking — for failure-mode focus, inversion is more efficient.

Rule of thumb: Spend analysis proportional to reversibility × stakes. Low-stakes + reversible = first-order only.

The Protocol: Cascading Consequence Analysis

Simplified Mode (Quick Decisions)

Use when the decision is reversible and low-stakes (Bezos Type 2 / Farnam Street "two-way door": you can walk back through the decision cheaply). Source: Farnam Street, "Reversible and Irreversible Decisions."

  1. What happens first?
  2. And then what?
  3. And then what? Stop when effects become speculative or plateau.

Escalate to Full Protocol if any of these apply: (a) the decision is irreversible or hard to reverse (Bezos Type 1 / one-way door), (b) the stakes are high enough that a missed cascade causes material harm, or (c) a Simple Mode pass reveals a feedback loop or non-obvious incentive that needs tracing.

Full Protocol (Important Decisions)

Step 1: First-Order Effects

Identify the immediate outcomes. Question: "What happens first?"

Step 2: Second-Order Effects

For each first-order effect, ask: "And then what?" Trace the chain forward.

Step 3: Third+ Order Effects

Continue: "And then what happens as a result of THAT?"

Step 4: Time Horizon Analysis

For each effect, estimate when it manifests:

TimeQuestionCalibration
ImmediateWhat happens in 10 min/hours?Halve your time-to-effect estimate. (How soon, not how big.)
Short-termWhat happens in 10 days/weeks?Halve your time-to-effect estimate.
Medium-termWhat happens in 10 months/years?Halve your time-to-effect estimate.
Long-termWhat happens in 10+ years?Quarter your time-to-effect estimate.

Calibration warning (Farnam Street, Shane Parrish): the halving rule applies to when an effect manifests, not to how strong it will be. Systematic optimism is the default human bias on time horizons — people under-estimate how long change takes, especially cumulative or institutional change. Source: Farnam Street, "Second-Order Thinking." Do not treat 30 min or 5 yrs as precise; they are anchors to deflate optimism.

Step 5: Feedback Loop Detection

Connect to systems-thinking for loop analysis.

  • Reinforcing loops: Success → resources → more success (or failure → learned helplessness → worse failure)
  • Balancing loops: Growth → limits → slower growth

Heuristic: Does a second-order effect create conditions that amplify the original? If yes, you have a reinforcing loop.

Stop Conditions

Stop tracing chains when:

  • Effects plateau — further steps show diminishing returns
  • Effects become speculative — no mechanism, just imagination
  • Decision stakes don't justify deeper analysis
  • You've traced through one full reinforcing or balancing loop
  • Time spent exceeds value of further certainty

Pareto Guidance: 80% of long-term impact comes from:

  1. Feedback loops that amplify or dampen
  2. Effects that change incentives
  3. Effects that affect future options

Deprioritize: linear chains without loops, effects that plateau quickly, effects on inconsequential actors.

Second-Order Effects Check (The Missing Piece)

This skill ironically lacks explicit second-order self-analysis. Add this:

Ask: "What are the second-order effects of thinking second-order?"

  • Second-order: Analysis paralysis, missed opportunities, social friction
  • Third-order: You become known as someone who "thinks too much"
  • Fourth-order: You miss the second-order benefit of decisive action — sometimes making a decision creates learning no analysis could replace

Logic Check

Before accepting any chain, ask:

  1. Is there a documented mechanism linking A→B, or just correlation?
  2. Do actors in the chain have incentives that might break the link?
  3. Does this assume a stable environment?

Trace the mechanism, not just the direction.

Key Distinction: Second-Order vs Inversion

AspectInversion-ThinkingSecond-Order Thinking
Question"How could this fail?""And then what?"
FocusPrevent negative outcomesAnticipate cascading effects
ScopeFailure modesAll consequences (positive + negative)
TimeStatic analysisDynamic across time horizons
Useful forSecurity, safety, debuggingStrategy, planning, decisions

Combined Protocol: Run inversion to identify failure modes, then run second-order to trace what happens after those failures.

Skill Integration Matrix

SkillWhen to Chain
inversion-thinkingFor failure-mode focus — combine with second-order for cascade tracing
systems-thinkingFor feedback loop analysis — especially Step 5
decision-matrixBefore scoring criteria, run time horizon analysis on each
chestertons-fenceBefore concluding — verify you understand why cascade terminates or loops
first-principles-thinkingTo challenge the assumption that cascade analysis is necessary

Self-Improvement Protocol

After a decision plays out, log only meaningful misses:

Log to .learnings/CORRECTIONS.md:

- [YYYY-MM-DD] {decision} → {missed effect} ({actual outcome})

Log if:

  • Missed a cascade entirely
  • Time horizon was systematically wrong
  • Chain broke due to actor incentive
  • Feedback loop behaved differently than predicted

Do NOT log routine confirmations or obvious cascades.

Common Patterns

The Cobra Effect

Punishment accidentally creates incentives for the opposite behavior.

The Tolerance Paradox

Tolerating intolerance leads to its spread.

The Knowledge Curse

Teaching something reduces ability to remember not knowing it.

Acceleration Effects

Early wins create conditions for later problems (success → reduced caution → failure).

Option Limitation

Each decision closes future options — track what's being foreclosed.

Evaluations

Eval 1: Time Horizon Calibration

Scenario: User asks "What happens if we switch to monthly releases?" Expected: Traces immediate (release cadence change) -> short-term (dev workflow) -> medium-term (team habits) -> long-term (culture change). Uses halving heuristic. Pass criteria: Applies time horizon halving, does NOT assume immediate effects persist indefinitely.

Eval 2: Feedback Loop Connection

Scenario: User asks about "early success in new market." Expected: Identifies reinforcing loop: Success -> Resources -> More Success OR Success -> Reduced caution -> Failure. Pass criteria: Connects to systems-thinking, identifies which loop type applies.

Eval 3: Second-Order Self-Analysis

Scenario: Skill is being used on every minor decision. Expected: Applies meta-inversion — identifies analysis paralysis as second-order cost, suggests limiting use to reversible high-stakes decisions. Pass criteria: Recognizes over-use pattern, recommends proportional analysis investment.

Resources

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.