"You've got to have models in your head and array your experience, both vicarious and direct, on this latticework of models."
Charlie Munger
The name comes from Munger's idea: no single framework explains much on its own, but held together they compound into judgment. This repo is that latticework: one model per skill, most useful when you reach for several at once.
A collection of structured thinking and analysis frameworks, packaged as agent skills. Each skill lives in its own folder and is written to be model- and harness-agnostic, usable from Claude Code, Codex, Copilot CLI, Gemini CLI, or any agent runtime that can load a skill.
Most are business strategy, analysis, and management frameworks: SWOT, Porter's Five Forces, the Business Model Canvas, Hoshin Kanri, and so on. Others cover the human side of work: team development, behavior change, conflict, and negotiation. And a few are general-purpose thinking and research tools, including Cynefin, the Six Thinking Hats, a cognitive bias audit, the pre-mortem, MECE decomposition, the SCQA pyramid, STORM research, a key assumptions check, Fermi estimation, reference-class forecasting, and the OCEAN personality test.
Jump to a category:
- 🎯 Strategy and business analysis
- 💡 Process and innovation methods
- 🚀 Execution and change
- 🤝 People and teams
- 🧭 Thinking and decision
- 🔬 Research
- ✍️ Writing and communication
- 🧠 Personal insight
The classic business toolkit. Frame the problem, read the competitive environment, look inward at the organization, then decide where to grow and what to prioritize.
| Skill | What it does |
|---|---|
| tosca-problem-definition | Sharpens a fuzzy problem into a crisp, solvable problem statement using the TOSCA framework (Trouble, Owner, Success criteria, Constraints, Actors). A guided interview that challenges weak answers and stress-tests the framing with parallel critics. |
| mece-decomposition | Breaks a problem or metric into a MECE issue tree (no overlaps, no gaps). Generates competing top-level cuts in parallel, picks the sharpest, builds the tree, then audits it for mutual exclusivity and exhaustiveness with parallel checkers. |
| scqa-pyramid | Structures a recommendation so the answer comes first, using SCQA and Minto's Pyramid Principle. Generates competing framings in parallel, picks the sharpest, builds the supporting pyramid, then audits its logic and key line. Turns analysis into something a reader can follow. |
| strategy-kernel | Builds a real strategy from Rumelt's kernel: a diagnosis that names the crux, a guiding policy that makes a hard choice, and coherent action that carries it out. Parallel diagnosticians compete to name the crux, then synthesis crafts the kernel and audits it against the four hallmarks of bad strategy. Where SWOT and Five Forces diagnose, this one decides. |
| Skill | What it does |
|---|---|
| swot-analysis | SWOT analysis anchored to an objective. Fans out four parallel analyst subagents (one per quadrant), prioritizes the factors, then crosses them into a TOWS strategy matrix with ranked recommendations. |
| pestel-analysis | PESTEL macro-environment scan. Fans out six parallel analyst subagents (Political, Economic, Social, Technological, Environmental, Legal), then prioritizes the forces by impact and certainty and draws implications. |
| porters-five-forces | Porter's Five Forces industry analysis. Fans out five parallel analyst subagents, scores each force against its determinants with evidence, and synthesizes industry attractiveness and strategy. |
| scenario-planning | Builds several plausible futures instead of one forecast. Gathers driving forces, picks the two critical uncertainties as axes, develops the four resulting scenarios with one parallel subagent each, then derives strategy that holds across them and the signposts to watch. |
| blue-ocean-strategy | Blue Ocean Strategy and value innovation. Maps the industry's strategy canvas, then fans out four parallel subagents (Eliminate, Reduce, Raise, Create) to build a value curve that lifts buyer value and cuts cost at once, tested for focus and divergence. |
| Skill | What it does |
|---|---|
| value-chain-analysis | Porter's Value Chain analysis. One parallel analyst per activity assesses cost and value versus competitors, then synthesizes the sources of advantage, the value leaks, and the linkages. |
| mckinsey-7s | McKinsey 7-S organizational analysis. Fans out seven parallel analyst subagents (Strategy, Structure, Systems, Shared Values, Skills, Style, Staff), then runs an alignment analysis to find where the elements conflict and recommend fixes. |
| digital-maturity-assessment | Scores digital maturity across seven dimensions on a five-level scale with parallel assessors, then maps the binding gaps and a sequenced roadmap toward a target. |
| vrio-analysis | Tests whether a firm's resources are a source of sustained advantage (Valuable, Rare, Costly to Imitate, Organized). One parallel analyst per resource runs the four gates with evidence, then synthesis names the few that yield durable advantage. Complements value-chain analysis. |
| Skill | What it does |
|---|---|
| bcg-matrix | BCG Growth-Share Matrix portfolio analysis. Places each business unit (Stars, Cash Cows, Question Marks, Dogs) with one parallel analyst per unit, then assesses portfolio cash-flow balance and reallocation. |
| ansoff-matrix | Ansoff Matrix growth strategy. Fans out four parallel strategist subagents (Penetration, Market Development, Product Development, Diversification) to generate concrete options with risk, then sequences a prioritized growth path. |
| three-horizons | McKinsey Three Horizons model. Three parallel analysts place initiatives by maturity (core, emerging, options), then assess portfolio balance, pipeline gaps, and how each horizon should be funded and measured. |
| Skill | What it does |
|---|---|
| impact-feasibility-matrix | Prioritizes a list of options by impact and feasibility (the impact-effort matrix). Calibrates the axes, scores options in parallel, then sequences a plan of quick wins, major projects, fill-ins, and tasks to drop. |
| weighted-decision-matrix | Chooses among options against multiple weighted criteria. Calibrates the criteria, weights, and scale first, scores each option in parallel, then computes weighted totals, ranks, and runs a sensitivity check. Generalizes the impact-feasibility matrix to many criteria. |
| pre-mortem | Stress-tests a plan or decision before you commit. Fans out parallel failure-finder subagents across distinct risk lenses, scores the risks by likelihood and impact, and designs mitigations for the worst ones. |
| fmea | Failure Mode and Effects Analysis. One parallel analyst per function lists failure modes and scores Severity, Occurrence, and Detection, then synthesis ranks them by Risk Priority Number and recommends actions that cut S, O, or D. A scored counterpart to the pre-mortem. |
| rice-prioritization | Ranks a backlog by Intercom's RICE score (Reach x Impact x Confidence / Effort). Calibrates the scale, scores each in parallel, then ranks the results and runs a sensitivity check on confidence and effort. A product-roadmap version of the weighted decision matrix. |
Building the right thing, then building it well. The innovation and product tools find and shape what customers want; the process tools make the work that delivers it more reliable.
| Skill | What it does |
|---|---|
| design-thinking | Facilitates the five Design Thinking modes (Empathize, Define, Ideate, Prototype, Test) as an iterative loop, with parallel idea generation during Ideate and a living design document. |
| lean-startup | Runs a Build-Measure-Learn loop: surfaces leap-of-faith assumptions, ranks them by risk in parallel, designs a minimum viable test with a pre-committed metric, and decides pivot or persevere. |
| business-model-canvas | Maps how a business creates, delivers, and captures value across the nine blocks of Osterwalder's Business Model Canvas. One parallel analyst per block, then a synthesis pass checks whether the blocks cohere and flags the riskiest assumptions. |
| value-proposition-canvas | Osterwalder's Value Proposition Canvas, which zooms into customer-and-value fit. Six parallel analysts rank the customer profile (jobs, pains, gains) and the value map (products, pain relievers, gain creators), then a fit pass checks which top pains and gains the offer actually addresses. |
| jobs-to-be-done | Finds the job a customer hires a product to do. Fans out parallel analysts across the functional, emotional, and social job and the four forces of progress, then writes the job story and ranks outcomes by opportunity to show where solutions underserve. |
| kano-model | Classifies product features by how they drive satisfaction (must-be, performance, attractive, indifferent, reverse). One parallel analyst per feature works the functional and dysfunctional responses, then synthesis covers the must-bes first, picks performance features to compete on, and chooses a few delighters. |
| Skill | What it does |
|---|---|
| six-sigma-dmaic | Guides a Six Sigma DMAIC cycle (Define, Measure, Analyze, Improve, Control) with phase gates, parallel root-cause hypotheses verified against data, and a project document. |
| pdca-cycle | Guides a lightweight PDCA loop (Plan, Do, Check, Act): plan a change with a prediction, try it small, check against the prediction, and adopt, adjust, or abandon, then loop. |
| root-cause-analysis | Finds the root cause with the Ishikawa fishbone and 5 Whys. Parallel finders hunt causes across the 6 Ms, then drill the strongest with why-chains to verified roots and turn them into countermeasures that remove the root, not the symptom. |
Turning strategy into action and making it stick. Set and measure the goals, then lead the people through the change those goals require.
| Skill | What it does |
|---|---|
| balanced-scorecard | Translates a strategy into objectives and measures across the four Balanced Scorecard perspectives (Financial, Customer, Internal Process, Learning and Growth). One parallel analyst per perspective, then a synthesis pass links them into a cause-and-effect strategy map. |
| okr | Turns strategy into measurable goals with OKRs. Guides a drafting cycle (a qualitative objective, then three to five measurable key results), stress-tests the key results with parallel critics for outcome-versus-task, measurability, coverage, and ambition, and sets cadence and scoring. |
| hoshin-kanri | Deploys a strategy with Hoshin Kanri (policy deployment). Holds the vital few breakthrough objectives, fans out one strategist per annual objective to set priorities, targets, and owners, then builds the X-matrix, traces the golden thread, and sets the review cadence. |
| Skill | What it does |
|---|---|
| stakeholder-analysis | Maps the people and groups behind a project or change on Mendelow's power-interest grid. One parallel analyst per stakeholder assesses power, interest, and stance, then a synthesis pass builds the engagement plan and the coalition-and-blocker map. |
| kotter-change | Leads an organizational change through Kotter's eight steps (urgency, coalition, vision, communication, removing barriers, short-term wins, sustaining, anchoring). A phased facilitator with a gate check per step and a living change plan. |
| adkar | Diagnoses the people side of a change with ADKAR (Awareness, Desire, Knowledge, Ability, Reinforcement). Five parallel subagents assess each element per affected group, then a synthesis pass finds the barrier point where change stalls and designs targeted interventions. |
| force-field-analysis | Lewin's force field analysis for a proposed change. Parallel finders hunt the driving and restraining forces across several lenses and score each, then synthesis builds the net field and finds the highest-leverage moves, favoring the weakening of restrainers. |
Frameworks for the human side of work: how teams develop, how people respond to change, and how to handle conflict and negotiation.
| Skill | What it does |
|---|---|
| five-dysfunctions-of-a-team | Diagnoses a team against Lencioni's pyramid (trust, conflict, commitment, accountability, results). One parallel analyst per layer scores it against the tells, then synthesis finds the lowest broken layer, traces how a cracked base surfaces higher up, and sequences the rebuild. |
| psychological-safety | Diagnoses whether a team is safe for interpersonal risk-taking (Edmondson). Parallel assessors score the climate across voice, response to mistakes, help-seeking, inclusion, and challenge, then synthesis places the team on the safety-by-accountability grid and prescribes leader practices to reach the learning zone. |
| tuckman | Locates a team's stage of development (Forming, Storming, Norming, Performing, Adjourning). One parallel assessor per stage scores the team against its markers, then synthesis names the current stage and the leadership moves to advance, especially past the Storming stall. |
| Skill | What it does |
|---|---|
| com-b | Diagnoses why a behavior is or isn't happening (Capability, Opportunity, Motivation) via the Behaviour Change Wheel. Six parallel analysts test the sub-components, then synthesis names the binding constraint, guarding the reflex to blame motivation when the gap is really opportunity or capability. |
| scarf | Analyzes a change or message for social threat across five domains (Status, Certainty, Autonomy, Relatedness, Fairness). One parallel analyst per domain scores the threat and reward, then synthesis ranks the biggest threats and designs moves to lower threat and raise reward. |
| self-determination-theory | Diagnoses thin motivation by testing the three basic psychological needs (Autonomy, Competence, Relatedness) for both satisfaction and frustration. One parallel analyst per need, then synthesis names the thwarted need driving the problem, places the motivation on the controlled-to-autonomous continuum, and prescribes need-supportive moves. Guards against blaming generic low motivation when a specific need is being thwarted. |
| Skill | What it does |
|---|---|
| thomas-kilmann | Chooses the conflict-handling mode that fits a situation (Competing, Collaborating, Compromising, Avoiding, Accommodating) on the assertiveness-by-cooperativeness grid. One parallel analyst per mode scores its fit, then synthesis recommends a primary mode and a fallback and flags a default mismatched to the situation. |
| negotiation-prep | Prepares for a negotiation with principled, interest-based bargaining plus BATNA and ZOPA. Parallel analysts work both sides' interests, BATNAs, and options for mutual gain, then synthesis maps the ZOPA and sets target and walk-away. Interests over positions; your BATNA is your power. |
General-purpose reasoning tools, not tied to business strategy. Reach for these to pick the right approach to a problem or to examine a decision from every angle.
| Skill | What it does |
|---|---|
| cynefin | Sorts a messy situation's issues into Cynefin's domains (Clear, Complicated, Complex, Chaotic, Disorder). Classifies each in parallel, maps the portfolio across the domains, flags the Clear-Chaotic cliff, and routes each to its fitting response. A meta-tool for picking the right tool. |
| six-thinking-hats | De Bono's Six Thinking Hats examines an idea from every angle. The Blue hat sets the focus question, five hats (White, Red, Black, Yellow, Green) examine it in parallel, then a Blue-hat synthesis reaches a conclusion with actions and risks. |
| cognitive-bias-audit | Audits a decision, plan, or forecast for the cognitive biases distorting it. One parallel hunter per bias family cites where the reasoning is exposed, then synthesis ranks the most dangerous and prescribes structural debiasing. Evidence from the actual reasoning, not generic warnings. |
| decision-tree-analysis | Maps a decision's choices and chance events into a tree, then folds back the probabilities and payoffs to the highest expected-value path. One parallel analyst per alternative estimates its branch, then synthesis computes the expected values, the break-even probabilities, and what resolving an uncertainty is worth. The quantitative-under-uncertainty counterpart to the weighted decision matrix. |
| ooda-loop | Runs Boyd's Observe-Orient-Decide-Act loop for a fast-moving or competitive situation, where cycling faster than the situation changes is the edge. A facilitator that spends its effort on Orient, breaking and rebuilding the mental model each loop, and tracks whether you are gaining or losing tempo. The fast-tempo counterpart to the deliberate PDCA loop. |
| first-principles | Rebuilds a problem from fundamental truths instead of by analogy to how it is usually done. Parallel subagents strip the inherited assumptions across distinct lenses to find the bedrock that survives, then synthesis reconstructs a solution from it and contrasts that with the conventional approach to show what convention was costing. |
| inversion | Attacks a goal backward: instead of how to succeed, asks what would guarantee failure, then turns those failure paths into things to avoid. Parallel finders generate failure modes across lenses, then synthesis ranks them and converts the worst into bright lines and constraints. Where the pre-mortem stress-tests a formed plan, inversion needs none. |
Investigating a question and pressure-testing what you find, from multi-source research to weighing rival explanations, checking assumptions, and estimating quantities you cannot look up.
| Skill | What it does |
|---|---|
| storm-research | PhD-level research on a topic. Fans out five expert perspectives as parallel subagents, maps their contradictions, synthesizes a briefing, and peer-reviews it with confidence scores. Parallel replication of Stanford's STORM method. |
| analysis-of-competing-hypotheses | Decides which of several rival explanations the evidence actually supports (Heuer's ACH). Proposes a near-exhaustive hypothesis set, scores every datum against each in parallel by working to disprove, drops non-diagnostic evidence, then ranks by fewest weighted inconsistencies. Ranks by disconfirmation, not confirmation. |
| key-assumptions-check | Surfaces the assumptions an analysis silently rests on (Heuer & Pherson's KAC). Parallel proposers elicit the full set, test each for confidence and failure conditions, then name the few linchpins. Where ACH tests the hypotheses, KAC tests what they rest on. |
| fermi-estimation | Produces an order-of-magnitude estimate with no data at hand (Fermi's method). Decomposes the unknown quantity in parallel into a chain of factors, estimates each as a range, multiplies the chain, then cross-checks against an independent decomposition. The inside-view counterpart to reference-class forecasting. |
| reference-class-forecasting | Forecasts a duration or cost from how similar past cases turned out (Kahneman's outside view). Builds a reference class in parallel and places the case in its base-rate distribution, checked for optimism. Counters the planning fallacy with base rates, not case specifics. |
Tools for the words themselves, not the analysis behind them. Make a draft land in fewer words and without the marks of machine prose.
| Skill | What it does |
|---|---|
| concise-writing | Writes or edits text so it is only as long as it needs to be and carries no signs of AI writing. Anchors to the purpose and reader, runs the text down a cut ladder (whole sections before words), rewrites the AI tells, and reads it aloud, looping draft to audit to final. Guards against over-trimming: the floor is meaning, not word count. |
Self-assessment questionnaires. Answer the statements and get a scored, written read on your personality.
| Skill | What it does |
|---|---|
| ocean-personality-test | Conducts the Big Five (OCEAN) personality test conversationally using the public-domain 50-item IPIP questionnaire, scores each dimension, and writes a detailed markdown report. |
| hexaco-personality-test | Conducts the HEXACO six-factor personality test conversationally using the public-domain 240-item IPIP-HEXACO questionnaire, scoring each factor and its facets. Adds Honesty-Humility, the integrity dimension Big Five omits. A longer, facet-level alternative to the OCEAN test. |
You don't have to prepare documents or fill in a form. Invoke the skill and give it what it needs in plain language. It asks for anything missing. Inputs come in a few shapes:
- Just invoke it. Questionnaires interview you, no prep: ocean-personality-test, hexaco-personality-test.
- Give a subject, then get interviewed. Guided framing and facilitation that draws the rest out of you: tosca-problem-definition, okr, design-thinking, pdca-cycle, ooda-loop, strategy-kernel, decision-tree-analysis, self-determination-theory.
- Give a topic, or paste your reasoning. Research and thinking tools that work on what you hand them: storm-research, cognitive-bias-audit, pre-mortem, six-thinking-hats, first-principles, inversion.
- Give a subject and an objective. Most analyses; the objective is what anchors them: swot-analysis, pestel-analysis, porters-five-forces, business-model-canvas.
- Give a list of options to score. Prioritization and decision tools: rice-prioritization, weighted-decision-matrix, impact-feasibility-matrix.
- Hand over text to fix, or a brief to write. Say what it is for and who reads it: concise-writing.
Each skill's README has a How to get started note, and its references/example.md shows a full worked example: what to type, the input filled in, and what you get back.
A skill is a self-contained unit of reusable instructions an agent loads on demand. It tells the agent how to do something (a procedure, a discipline, a domain workflow) without hard-coding any one platform's tools or any specific model.
latticework/
├── README.md
├── LICENSE
├── .claude-plugin/
│ └── plugin.json # Claude Code plugin manifest
└── skills/
└── <skill-name>/
├── SKILL.md # the skill itself, loaded by the agent (required)
├── README.md # human-facing overview, rendered on GitHub (optional)
└── references/ # files the skill loads at runtime (optional)
One folder per skill under skills/, named after the skill (kebab-case). Three kinds of file can live in it, split by who reads them:
SKILL.md(required) is for the agent. It is the procedure the runtime loads on demand.README.md(optional) is for humans. GitHub renders it when someone opens the folder, so put the what and why here. When present it follows the README structure below.references/(optional) is also for the agent. It holds material the skill pulls in at runtime, such as prompt templates, scripts, or worked examples, including anexample.mdthat shows how the skill is invoked and what input it expects. It is not the place for human documentation.
Each SKILL.md carries YAML frontmatter:
---
name: <skill-name>
description: <one line, when an agent should reach for this skill>
---
<the skill body: the procedure, in plain action language>Every skill's README.md uses the same four sections, in this order, so the catalog reads consistently. The H1 is the skill name.
- What this is is the framework itself: what it is, its parts, and where it came from (founder, source, date). It ends with a link to the skill's framework reference file, if it has one.
- What it does is the skill's mechanics: how it runs (its stages and parallel subagents) and what report it produces, plus any guardrails that keep it honest.
- When to use it says when to reach for the skill, and which sibling skill to use instead when the question is a better fit for one. Point to siblings with relative links.
- How to get started says what the user provides, then links
references/example.mdandSKILL.md.
See swot-analysis for the reference example.
- Harness-agnostic: speak in actions ("read the file", "dispatch a subagent", "run the tests"), never in one runtime's tool names. Let each platform map actions to its own tools.
- Model-agnostic: no dependence on a specific model, context window, or vendor feature.
- Self-contained: everything the skill needs lives in its folder.
- Single responsibility: one skill, one job. Compose, don't bloat.
Install the skills with skills, a small CLI that pulls SKILL.md folders from GitHub and drops them into whichever agents you have (Claude Code, Codex, Cursor, Copilot, and more):
# see what's in this repo
npx skills add l4ci/latticework --list
# add one skill
npx skills add l4ci/latticework --skill swot-analysis
# or add all of them
npx skills add l4ci/latticework --allStandalone (any runtime): or copy a folder from skills/ into your agent's skills directory by hand, or point your runtime at skills/. Conventions vary by platform:
- Claude Code: drop a
skills/<name>/folder under~/.claude/skills/and invoke via the skill mechanism. - Codex / Copilot CLI / Gemini CLI: place under the platform's skills path; most auto-discover by folder.
Check your runtime's docs for the exact path.
New skill → new folder under skills/ with a SKILL.md. Keep it harness-agnostic, keep it focused, and keep supporting material inside the folder. Ship a README.md that follows the README structure: the four sections in order, with sibling cross-references in When to use it and no em dashes.
Full walkthrough, including the SKILL.md anatomy, the references/ files, the catalog entries to update here, and a pre-PR checklist, is in CONTRIBUTING.md.