Consulting Research to Output

A hypothesis-led consulting workflow for turning an ambiguous brief into a defensible answer and a finished memo, analysis, HTML deck, or editable PowerPoint.

The suite treats slides as the final expression layer—not the consulting work itself. One central case lead clarifies the real question, decomposes it through a business mechanism, forms falsifiable rival hypotheses, directs evidence collection and analysis, adjudicates what the facts establish, and only then approves the storyline and output.

This is an original, non-branded skill suite. It is not affiliated with or derived from any consulting firm's proprietary methodology, template, or trademarked material. The public package contains only original instructions and synthetic demo assets—no private source file, extracted slide content, client material, or firm-branded asset.

Help improve the workflow: If an answer felt shallow, the reasoning broke, or a page was hard to understand, report the specific failure. A redacted screenshot or one problematic page is welcome. You can also ask your agent to prepare and submit the feedback.

Install

Install the full suite:

npx skills add Rkamirage/consulting-research-to-output

Or select only the skills you need when the installer prompts you.

Start here

For an ambiguous consulting problem, invoke the central skill:

Use $run-consulting-research-to-output.

Before research, help me clarify the real question and the missing context.
Then decompose it into mechanism-based subquestions and falsifiable rival
hypotheses, test the evidence, analyze any supplied data, adjudicate the answer,
and finish the requested deliverable.

Brief: [your problem]
Available files/data: [what you have]
Audience and output: [if known]
Current belief, rival, and access boundary: [if known]

If material context is missing, the central lead asks only the few high-information questions that can change the work. It does not force a long intake form.

The six skills

Skill	Owns	Does not own
`run-consulting-research-to-output`	Question framing, MECE decomposition, rival hypotheses, dispatch, adjudication, answer revision, and completion	Specialist evidence collection or slide composition
`build-consulting-evidence-base`	Bounded research, original-source pursuit, counterevidence, conflicts, sufficiency, and missing-data requests	Final recommendation or storyline
`execute-consulting-analysis`	Calculations, models, benchmarks, scenarios, checks, sensitivity, and analytical limits	Overall hypothesis adjudication
`structure-consulting-storyline`	Belief journey, title spine, and concise page briefs with proof and sources	Layout and production
`communicate-consulting-message`	Message family, claim-state language, audience adaptation, and the appropriate close	Multi-page sequencing or visual design
`design-consulting-slides`	Art direction, visual models, HTML/PPTX production, rendered QA, and the finished deck	Rewriting unsupported conclusions

The modules are capabilities, not a mandatory six-agent assembly line. The central lead uses only what the unresolved problem needs and avoids per-page subagents.

Operating model

Clarify the real question and reader
        ↓
Choose a consulting identity and business mechanism
        ↓
Build MECE subquestions and rival, falsifiable hypotheses
        ↓
Specify proof / refutation / inconclusive conditions
        ↓
Research original evidence and analyze supplied data
        ↓
Central lead adjudicates each hypothesis and revises the answer
        ↓
Build a detailed content spine
        ↓
Approve storyline and message
        ↓
Produce the requested artifact and review the actual render

Research returns evidence, not a recommendation. Analysis returns a checked result, not the overall answer. Storyline and slide work begin only after central adjudication.

What “good” looks like

The issue tree follows how the business outcome works, not a generic market/customer/competitor checklist.
Each priority hypothesis is a complete statement that can be wrong and has a credible rival.
Search results are discovery leads; decision claims trace to opened and read original sources.
Vendor claims do not become proof of customer demand, realized value, or superiority.
Missing user-controlled data becomes a small, executable access request, including what a no-data answer can and cannot conclude.
The answer states observed facts, analytical inferences, unresolved uncertainty, and the decision boundary separately.
A deck page carries a decisive fact, number, mechanism, worked example, or explicit gap—not an empty framework waiting to be populated.
Rendered comprehension and source reconciliation matter more than clean boxes or object-tree validity.

How effort scales

There is no work-level selector. The central lead derives the required effort from decision impact, uncertainty, testability, reversibility, evidence availability, and dependency. It expands research only while another feasible route could change the answer, and narrows the released conclusion when decisive evidence remains unavailable.

Formal audit, regulatory, permission, or release controls are added only when the engagement actually requires them.

Examples and evaluation

See case-prompts.md for end-to-end prompts and observable acceptance criteria. They are designed to expose shallow research, generic decomposition, premature storyline creation, and text-block slide design.

See the public consulting-skill comparison case for a complete English research memo, before/after execution trace, and rendered PDF deck built with this suite.

Run the public-package checks locally:

python3 scripts/validate_public_package.py

For maintainers, ~/.codex/skills/ may be the development workbench while this repository remains the canonical published projection. Project only the intended public-managed files, require parity with python3 scripts/sync_local_install.py --check, and run python3 scripts/release_check.py before pushing. Use --install only to restore the published projection locally. See CONTRIBUTING.md for the full drift-safe workflow.

Feedback and responsible use

The most useful contribution is a specific failure, not a generic rating. Use the output feedback form to report:

a weak question decomposition or hypothesis;
shallow, conflicting, or unsupported research;
an analysis or conclusion that does not hold;
a storyline or message that does not connect;
a problematic page, visual, or rendering defect; or
an installation, routing, or usability problem.

You may drag a redacted screenshot or one problematic page into the issue form. If you prefer, follow the agent-assisted feedback instructions so your agent can prepare a minimal report and submit it with GitHub CLI.

Remove confidential, personal, proprietary, copyrighted, or client-identifying information before sharing. Do not upload a complete source deck or dataset when one excerpt can demonstrate the problem.

The skills do not create authorization to access private data, contact people, spend money, publish claims, or use a client's brand. Apply professional judgment and the permissions of your environment.

License

MIT

Consulting Research To Output

At a glance

README