58 consulting methods your AI agent can work through, step by step
Give your agent a working guide for the question in front of you.
30-second start · Everyday uses · Choose a skill · Browse all 58 · Examples · Quality · 中文 · O2V Framework
An AI answer can sound sensible and still leave you unsure what to do next. When the notes are scattered and people disagree about the cause, a recommendation needs supporting work you can examine.
Consulting Skills starts there. The idea behind the project is to turn established consulting methods into working guides an agent can follow. The guide connects the material you provide to a concrete output, such as an evidence ledger or a decision table, and explains how to check it.
Asking an agent to “use SWOT” leaves much of the execution unspecified. A skill fills in those details. It sets out the evidence needed and the steps to work through, with boundaries for situations where the method does not fit.
Li Zhi initiated this library to make consulting methods easier to apply through AI agents. It contains 58 standalone skills. You can start with one relevant to your question and add another if its output would help you make the decision.
Imagine a team deciding how to improve use of an internal request tool. The input and output below are a synthetic illustration of the format. They are not a customer story or a model test result.
Give the agent these notes:
Use evidence-map to help decide whether we should add training this week.
A. Dashboard export: weekly active users fell from 42 to 28 across two weeks.
B. Interview notes: 3 of 5 people said they could not see their request status.
C. Sponsor's explanation: “People just need more training.”
The decision is due Friday. Separate evidence, explanations and unknowns.
A useful output would make the decision easier to examine:
| Input | What it supports | What it does not establish |
|---|---|---|
| A: dashboard export | Reported activity fell in this two-week window | Why it fell; whether the weeks are comparable |
| B: five interview notes | Status visibility is a plausible problem for some interviewees | How common the problem is, or whether it caused lower use |
| C: sponsor's explanation | A training hypothesis to investigate | Evidence that training is the cause or the right intervention |
The accompanying evidence map would rate source strength and relevance, record the missing information, and finish with a small next move:
Do not conclude yet: the notes do not establish that a training gap caused the decline. Proposed check before Friday: an analyst, once assigned, reviews whether the two dashboard weeks are comparable and follows up on the five interviews. Evidence of users not knowing how to complete a request would support the training hypothesis; evidence of users completing requests but being unable to track them would support testing status visibility. Neither finding alone establishes population-wide impact.
Now there is something to review and a test to run. The full evidence-map guide adds the evidence ledger, confidence readout, proposed owners and deadlines, and the result that would change the recommendation. Missing evidence stays visible.
The work goes beyond collecting 58 framework names. Each guide has explicit inputs, use boundaries, execution steps, an output template and a quality gate. Method-specific differences matter: a scoring method needs valid scales and denominators; a causal diagnosis needs competing explanations and tests; a stakeholder map must keep unknown influence visible.
The library was audited across all 58 methods for those contracts. Findings led to focused refinements in 13 methods, clearer worked examples and a separate service-plan review. The changelog records the scope. Structure and links are checked by the repository's validation scripts; actual model outputs need separate evaluation.
For the audited skill source, a recorded paired, blinded, order-reversed evaluation of the 70 representative cases produced 65 PASS and 5 TIE/INCONCLUSIVE under the published artifact rubric. All cases remain in the denominator. These are AI reviewer judgments; exact runtime model metadata was not exposed. They do not establish independent human expert endorsement, real business outcomes or universal superiority. The raw records and limitations are retained in the audit delivery; this README does not claim a public evidence download. The 24 method-selection cases are available but were not executed in that 70-case batch.
The guides are meant to be used and questioned. See how quality is checked for what the different checks establish.
Install the library with the open Skills CLI:
npx skills add fzfclee/consulting-skillsOr install one method with GitHub CLI:
gh skill install fzfclee/consulting-skills systems-thinking --agent codex --scope userThen ask your agent:
Use systems-thinking to diagnose why this issue keeps returning after repeated fixes.
Separate evidence from assumptions. Identify feedback loops, delays, leverage points,
side effects, and the next validation signal.
Every skill follows the same basic pattern:
Business question
↓
Check inputs and use boundary
↓
Execute explicit method steps
↓
Produce a reusable decision artifact
↓
Run the method-specific quality gate
You do not need to be a consultant or memorize a framework. Give your agent the situation, the facts you have, what you are unsure about, and the decision you need to make. Then start with one relevant skill.
| A question you might have | Useful starting skills | What you get back |
|---|---|---|
| Should I stay in this job, take an offer, or keep looking? | evidence-map, scenario-planning |
Facts and assumptions, plausible futures, and conditions for changing course |
| Which product, supplier, or large purchase should I choose? | decision-matrix, cost-benefit-analysis |
A comparison that shows tradeoffs, weak evidence, and hard constraints |
| Why does the same problem keep coming back at work? | change-event-timeline, systems-thinking |
A timeline, recurring pattern, feedback loops, and a safer intervention |
| Does a side business or new product make economic sense? | business-model-canvas, break-even-analysis |
Key assumptions, the break-even threshold, and what to test before spending more |
| Why are people not using a product or internal tool? | jobs-to-be-done, metrics-tree |
Adoption hypotheses, missing evidence, actions, and success measures |
| How do I get a proposal through a complicated group? | stakeholder-power-map, communications-plan |
Decision makers, informal influence, engagement moves, and a communication plan |
Use one method first. Add a second only when it answers a different question that could change the decision.
For an existing process-improvement proposal, provide the versioned original text, scoped evidence and rules, constraints, approval status and available operating data. Ask for an Evidence Map, a Deductive Reasoning check and a Validation Plan only where each is needed. Expect a traceable issue list, impact caveats, minimal wording revisions and pass/adjust/stop conditions. Missing evidence or approvals remain unresolved. The synthetic service-plan review demonstrates this artifact; it is authored guidance, not a native execution test or proof of business effectiveness.
Start with the problem you have, not the framework you remember.
| Situation | Start with | What it should produce |
|---|---|---|
| Facts, opinions, and assumptions are mixed | evidence-map |
Evidence ledger and confidence gaps |
| A broad problem needs a clean structure | issue-tree |
Decision-oriented problem tree |
| A problem keeps returning | systems-thinking |
Loops, delays, leverage points, side effects |
| Several causes remain plausible | abductive-reasoning |
Competing explanations and discriminating tests |
| Multiple options need comparison | decision-matrix |
Criteria, tradeoffs, sensitivity, recommendation |
| Stakeholders hold hidden veto power | stakeholder-power-map |
Influence map and engagement moves |
| Product adoption or realized value is weak | jobs-to-be-done |
Demand-side diagnosis and adoption hypotheses |
| A plan needs a failure rehearsal | pre-mortem |
Failure modes, prevention, triggers, contingencies |
| The future depends on critical unknowns | scenario-planning |
Robust moves, contingent moves, signposts |
| Work is blocked by one limiting factor | constraint-analysis |
Constraint proof and exploitation plan |
If the right starting point is still unclear, browse catalog.yaml or look at the seven controlled comparisons.
| Method family | Skills | Typical decisions |
|---|---|---|
| Problem framing and evidence | 8 | What is true, missing, noisy, or poorly framed? |
| Reasoning and root cause | 8 | Which explanation is strongest, and why? |
| Systems, risk, and futures | 5 | What could persist, fail, or change over time? |
| Strategy, market, and commercial | 10 | Where to play, how to compete, and how to win? |
| Customer, product, and experience | 7 | What do users need, adopt, value, or reject? |
| Prioritization and economics | 7 | Which option deserves time, money, or sequence priority? |
| Measurement and performance | 3 | What should be measured and governed? |
| Stakeholder, change, and governance | 7 | Who decides, influences, owns, or resists? |
| Execution and validation | 3 | What is the smallest credible action and proof? |
Browse all 58 skills by method family
5w1h-analysis ·
affinity-diagram ·
assumption-inventory ·
evidence-map ·
issue-tree ·
mece-framework ·
mind-map-analysis ·
signal-vs-noise-filter
abductive-reasoning ·
constraint-analysis ·
deductive-reasoning ·
first-principles-thinking ·
fishbone-diagram ·
five-whys-root-cause ·
hypothesis-tree ·
inductive-reasoning
critical-uncertainties ·
pre-mortem ·
risk-matrix ·
scenario-planning ·
systems-thinking
account-plan ·
business-model-canvas ·
competitive-positioning ·
deal-strategy-map ·
go-to-market-diagnosis ·
pestel-analysis ·
porter-five-forces ·
pricing-strategy-check ·
swot-analysis ·
win-loss-review
customer-segmentation ·
customer-success-health-score ·
empathy-map ·
jobs-to-be-done ·
kano-model ·
service-blueprint ·
user-journey-mapping
break-even-analysis ·
cost-benefit-analysis ·
decision-matrix ·
effort-impact-matrix ·
rice-scoring ·
weighted-scorecard ·
wsjf-prioritization
balanced-scorecard ·
metrics-tree ·
north-star-metric
change-event-timeline ·
change-impact-analysis ·
communications-plan ·
force-field-analysis ·
power-interest-grid ·
raci-matrix ·
stakeholder-power-map
Every skill must include five practical sections:
Required InputsWhen Not To UseStep-by-Step ExecutionOutput TemplateQuality Gate
The repository checks those sections automatically:
| Check | Current coverage |
|---|---|
| Standalone skills validated | 58 |
| Evaluation prompts covering all skills | 70 |
| Method-selection discrimination cases | 24 |
| Controlled comparisons | 7 |
| Neutral catalog entries | 58 |
| Portability or local-path failures allowed | 0 |
The validator checks names, frontmatter, required sections, links, UTF-8 text, catalog consistency, and portability. These structural checks catch packaging errors and missing contract sections; they do not establish semantic quality or prove that a model will make the right decision. The evaluation contract therefore publishes the prompts and scoring rubric separately.
| Decision problem | Example route |
|---|---|
| Clarify an ambiguous problem | Evidence Map → Issue Tree |
| Diagnose a recurring issue | Timeline → Systems Thinking → Constraint Analysis |
| Prioritize competing options | Decision Matrix → Risk Matrix |
| Build a stakeholder strategy | Stakeholder Power Map → Account Plan |
| Improve product adoption and ROI | Evidence Map → JTBD → Metrics Tree |
| Assess market entry | PESTEL → Five Forces → Positioning |
| Evaluate a career decision | No method vs Evidence Map → Stakeholder Power Map → Scenario Planning → Weighted Scorecard |
These authored illustrations are not a universal router, native execution receipts or a scientific model benchmark. Each presents the same input in two forms: a representative direct answer with no named method, and a worked answer using the listed method chain. The comparison shows the evidence boundary, method workpapers, handoff between steps, final decision artifact, action plan, success measures, and reversal conditions.
In the seven comparisons, the representative direct answer often reaches a sensible recommendation quickly. The worked method answer may reach the same conclusion. The difference is that it leaves behind workpapers, the handoff from one method to the next, owners and timing, success measures, decision gates, and triggers for changing course.
A method chain is intended to make the reasoning easier to review and the plan easier to run. It does not make the conclusion automatically correct.
| This repository is | This repository is not |
|---|---|
| A portable library of standalone consulting method skills | A claim of ownership over established methods |
| A set of execution instructions for AI agents | A substitute for evidence or professional judgment |
| A neutral catalog with explicit boundaries | A rule that every problem needs a framework |
| An open method resource in the O2V Framework ecosystem | The complete O2V Framework or CLEAR / Signal-to-Action methodology |
Consulting Skills is initiated by Li Zhi and maintained as an open method library in the broader O2V Framework knowledge ecosystem. O2V Framework and CLEAR / Signal-to-Action are maintained separately as complete methodology assets.
skills/<skill-name>/SKILL.md 58 standalone method skills
catalog.yaml neutral discovery metadata
examples/ seven controlled decision comparisons
evaluations/ behavior and method-selection test cases
scripts/ catalog generation and validation
Contributions should improve method fidelity, trigger clarity, evidence discipline, output usability, or decision impact. A new method should fill a real decision need rather than increase the count.
Read CONTRIBUTING.md, open an issue, or submit a pull request.
This project does not claim ownership of established consulting tools, analytical methods, academic theories, or management frameworks. Method names are used descriptively; their rights and academic contributions remain with their respective creators, researchers, institutions, and rights holders.
Apache License 2.0 applies to this repository's original skill text, executable structure, examples, catalog, and validation code. O2V Framework, CLEAR, Signal-to-Action, and their original framework expressions and assets are separate works by Li Zhi and are not licensed under this repository's Apache-2.0 license.
See NOTICE and ATTRIBUTIONS.md.
Official website: www.o2vframework.com
Email: contact@o2vframework.com
Found it useful? A star helps others find it. Found a weak method? Open an issue.