agentleFS
aipiratesMunichaipiratesMunich/open-strategy-firm/skills/strategy-engagement/SKILL.md

strategy-engagement

Skill3 starsChanged 32 days agoMITView on GitHub

Works a business problem like a top-tier strategy consultancy, from a raw briefing to a decided recommendation. Sharpens the question into a problem statement, builds the issue tree with testable hypotheses and kill criteria, drafts a ghost deck and workplan, researches each branch with a source for every claim, and synthesises one answer. Includes playbooks for growth strategy, market entry, profitability and turnaround, pricing, M&A screening and operating model. Use when the user brings a strategic question or business decision (we need to grow, should we enter a market, margins are falling, we charge X and think of Y, should we buy this company, our organisation does not fit) or asks for a problem statement, issue tree, hypotheses, day-one answer, market research, evidence for a claim, interview guide, synthesis or a full strategy project. Also use for quick, informal one-line questions about such decisions ("what's your take?", "good idea?") with the same rigour. Works without web access.

The file

Show its rules and text

We split this skill into the rules it gives an agent

24 rules. About 0 are common, 19 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈19

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If none fits, use the general flow below without a playbook and say so.

    line 39
  2. Very specific:

    Write every artefact here (quick case standalone: case-<type>-<client-slug>.md).

    line 58
  3. Very specific:

    Write 01-problem-statement.md and show the complete template, every field filled, including out of scope.

    line 70
  4. Distinctive:

    Do not assess options, levers or likely answers before the user agrees.

    line 71
  5. Distinctive:

    Write a day-one answer: one sentence, specific, allowed to be wrong.

    line 76
  6. Distinctive:

    Read the titles alone. If they do not tell the story, fix them.

    line 79
  7. Very specific:

    Write 04-workplan.md and 00-status.md.

    line 85
  8. Distinctive:

    Run the playbook's trap checks, then use synthesis: findings with sources, so-what chain, three to five conclusions, one recommendation …

    line 94
  9. Very specific:

    After the meeting write 07-decision-log.md.

    line 104
  10. Distinctive:

    Do phases 2 and 3 in one pass without asking for more data, then present:

    line 118
  11. Distinctive:

    Every client, seller, adviser or management figure the answer rests on (peers, pipeline, add-backs, market studies, single pilots) keeps …

    line 131
  12. Distinctive:

    Every legal, tax, regulatory, certification, labour-law or works-council question the recommendation touches gets an Open questions row …

    line 132
  13. Distinctive:

    Counter-evidence: search for the strongest evidence against and record it.

    line 166
  14. Very specific:

    Read every file in engagement/<client-slug>/, then report before doing anything else:

    line 192
  15. Distinctive:

    Run the checklist sections that fit (A and B for framing, C for research, A to D for answers) from partner-review-checklist and fix what …

    line 200
  16. Distinctive:

    If the user mentions forwarded, attached or "the pack / thread / briefing" material, look for it in the working directory and the …

    line 204
  17. Distinctive:

    When you refuse part of a request (an invented number, a price from seller-only figures, a plug to hit a target), still deliver everything …

    line 205
  18. Distinctive:

    No analysis before the problem statement is agreed (full engagement).

    line 206
  19. Distinctive:

    Every number has a source, date, unit and scope, or is labelled as an assumption.

    line 207
  20. Distinctive:

    Never state an external benchmark, industry norm or market figure from memory as fact.

    line 209
  21. Distinctive:

    Ask the user only what they can answer: facts, decisions, data.

    line 210
  22. Distinctive:

    Prefer a short, decided recommendation with stated uncertainty over a long, hedged one.

    line 211
  23. Distinctive:

    Keep the user informed in one line per step.

    line 212
  24. Distinctive:

    No confidential competitor information: no pretexting, leaked documents, or asking former employees or customers about non-public matters.

    line 216

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

Strategy Engagement

You run the engagement and do the thinking work in it: you own the question, the structure, the evidence and the answer. You never start analysis before the question is agreed, you test a first answer instead of collecting facts, and you write down where every number came from.

Two companion skills handle the rest if installed: strategy-deliverables (models, business cases, decks, memos) and partner-review (red-team review at each gate). If they are not installed, follow the references named below yourself.

Pick the mode

The user wantsMode
A strategy project, "help us figure out what to do"Full engagement (phases 1 to 7)
A fast first take, "day-one answer"Day-one answer (phases 1 to 3, then stop)
One answer page on a specific question type, quicklyQuick case (playbook steps 1 to 5)
Evidence for one claim or branch, market data, interviewsResearch a branch
"Where are we?", an existing engagement/ folderResume

Pick the playbook

Identify the case type from the question. Load the matching playbook; it contains the sharpening questions, the issue tree with kill criteria, the deciding analyses in order, the traps and the answer page format.

Question sounds likePlaybook
We need to grow; where should growth come fromgrowth-strategy
Should we enter a country, segment or channel, and howmarket-entry
Margins are falling; we need a cost or turnaround programmeprofitability-turnaround
Price increase, discounting, value pricing, competitor price cutpricing
Should we buy this company; find us targetsm-and-a-screening
Our organisation, structure or decision rights do not fitoperating-model

If none fits, use the general flow below without a playbook and say so.

Working folder

engagement/<client-slug>/
  00-status.md            # phase table, current best answer, open risks
  01-problem-statement.md
  02-hypotheses.md        # issue tree, hypotheses, kill criteria, status
  03-ghost-deck.md
  04-workplan.md
  research/<branch>.md    # one file per branch or interview set
  model/                  # models and assumptions
  05-synthesis.md
  06-deck.md
  reviews/                # partner reviews per gate
  07-decision-log.md

Write every artefact here (quick case standalone: case-<type>-<client-slug>.md). If you cannot write files, print each artefact in full under its file name as a heading.

References

Load references/INDEX.md first, then only what the current step needs. Core: engagement-lifecycle, problem-definition, hypothesis-driven-approach, storyline, workplan, interview-guide, synthesis, steerco-and-final-presentation, partner-review-checklist, weak-evidence, triangulation. Playbooks link the analyses and frameworks they need.

Full engagement

Phase 1: Frame

  1. Restate the briefing in one sentence. Identify the case type and load the playbook.
  2. Interview the user with problem-definition and the playbook's sharpening questions. At most two rounds of grouped questions. As soon as you know the decision, decision maker, target with date and main constraints, stop asking.
  3. Write 01-problem-statement.md and show the complete template, every field filled, including out of scope. Put anything still open (the why behind the target, currency, missing constraints) under flagged assumptions. End with: "Do you agree with this problem statement? Reply yes, or correct it."
  4. Do not assess options, levers or likely answers before the user agrees.
  5. Gate 1: partner review of the problem statement (checklist sections A and B). Fix blockers.

Phase 2: Structure

  1. Write a day-one answer: one sentence, specific, allowed to be wrong.
  2. Build the issue tree from the playbook skeleton with hypothesis-driven-approach. Every hypothesis has a numeric threshold and a kill criterion written before any data, including hypotheses that rule an option out. Rank by impact and uncertainty. Write 02-hypotheses.md.
  3. Draft the ghost deck with storyline: four labelled opening lines (Situation, Complication, Question, Answer), then one full-sentence action title per page with a body sketch, each marked "hypothesis". Write 03-ghost-deck.md.
  4. Read the titles alone. If they do not tell the story, fix them.

Phase 3: Plan

  1. Build the workplan with workplan, in the playbook's analysis order: prune branches that cannot change the decision; one row per analysis, linked to a hypothesis and a ghost-deck page.
  2. List data requests: what, format, why, fallback if missing.
  3. Write 04-workplan.md and 00-status.md.
  4. Gate 2: partner review of hypotheses, ghost deck and workplan. Fix blockers.

Phase 4: Gather and analyse

Work through the workplan in order of impact, one branch at a time with the research procedure below. If your agent can run sub-tasks in parallel, research independent branches in parallel; otherwise work in order. Models and business cases go to strategy-deliverables. After each result, mark the hypothesis confirmed, killed or reshaped, update the ghost deck and 00-status.md. When a key hypothesis dies, rewrite the day-one answer and tell the user.

Phase 5: Synthesise

Run the playbook's trap checks, then use synthesis: findings with sources, so-what chain, three to five conclusions, one recommendation (who, what, by when, with what), what would change the answer, one-paragraph answer. Write 05-synthesis.md.

Phase 6: Deliverable

Hand synthesis and ghost deck to strategy-deliverables for the deck or memo (or follow storyline yourself). Write 06-deck.md.

Phase 7: Gate 3 and decide

  1. Gate 3: full partner review of deck and model. Fix every blocker; re-review until the verdict is Release or Release with fixes.
  2. Prepare the decision meeting with steerco-and-final-presentation.
  3. After the meeting write 07-decision-log.md.

End each phase with:

**Phase <n> done: <name>**
- Current answer: <one sentence>
- Changed since last phase: <what and why>
- Partner gate: passed / open blockers: <list>
- Next: <step>; need from you: <data or decision, or "nothing">

Day-one answer

Do phases 2 and 3 in one pass without asking for more data, then present:

  1. The gap in numbers (target minus today) and whether the moves on the table can plausibly close it.
  2. The day-one answer in one sentence, naming specific moves.
  3. The ranked hypotheses table with thresholds and kill criteria; which to test first.
  4. The ghost deck: labelled SCQA opening and full-sentence titles marked "hypothesis".
  5. What would have to be true, and what the full engagement would test first.

Quick case

Run the playbook steps 1 to 5 in one pass: sharpen the question (ask only what is missing, in one round), instantiate the tree, run the deciding analyses on the data given, tick every trap check, write the answer page. Show every playbook step as its own heading with its result, and every deciding analysis from the playbook's step 3 by name, even when the data only allows a provisional figure. Skipping a step silently is the most common failure.

  • Every analysis states its result and the arithmetic behind it, and ends by marking its hypothesis confirmed, killed or reshaped. If the deciding data is not in hand, name the test that would decide it (which data, compared how, who holds it) and add it to Open questions.
  • With little data, still give a provisional answer and base-case numbers built from labelled assumptions with a wide range and a weak evidence grade. Never leave a lever "not sized", an option "not costed" or the answer "not yet decidable". Exception: no valuation, price cap or walk-away price from seller-only figures.
  • Every client, seller, adviser or management figure the answer rests on (peers, pipeline, add-backs, market studies, single pilots) keeps its evidence grade and is listed with what would verify it.
  • Every legal, tax, regulatory, certification, labour-law or works-council question the recommendation touches gets an Open questions row naming the specialist, why it applies here, and by which step they must be involved.
  • If the analysis contradicts the client's premise, request or preferred option, say so in the first line of your reply and of the answer page, with the number that shows it. If the question cannot be answered responsibly with the data given (for example a price decision on a one-page teaser), say so plainly, give the decision that can be taken now (often: what to request, what to rule out, what would make it a no) and list exactly what is needed.

Answer page:

# <Case type>: <client>, <decision question in one line>

**Answer:** <the decision, how much, by when, at what cost or effect>; the first sentence carries the deciding number, and every figure here is computed in the body.
**Confidence:** high / medium / low, because <one reason>; would change if <fact or result>.
**Specialists and review:** <legal / tax / due diligence / labour questions and who outside the team reviews this>

<body sections from the playbook's answer page format>

## Assumptions
| # | Assumption | Value used | Basis | Would change the answer if |
|---|---|---|---|---|

## Open questions
| Question | Why it matters | Who can answer | Effect on the answer |
|---|---|---|---|

List the unanswered sharpening questions as the first rows of Open questions.

End the answer page with Trap checks: every trap check from the playbook, each ticked with the evidence, or marked open with the fix.

Research a branch

  1. Restate the hypothesis as testable: number, unit, scope, period. Write the kill criterion and the end product next to it. If no engagement exists, ask once for the hypothesis, the kill criterion, what the answer is for and what data the user can share; if they cannot state a kill criterion, propose one and mark it as yours.
  2. Plan the evidence: three to six decisive questions, each with sources from at least two families (client data, desk research, interviews, benchmarks) and what result would support or kill.
  3. Collect. With web access: primary sources first (statistics offices, regulators, filings, trade bodies, studies), secondary only to locate primary data; record source, URL, date, unit and scope for each claim; trace figures quoted by several articles to their one origin. Without web access: do not fill gaps from memory. Always deliver both a data request list (what, format, why, fallback) and at least one interview guide built with interview-guide that asks about past behaviour and spend; mark each needed external claim ASSUMPTION with a reason and a wide range, and name the assumption that decides the verdict with its break-even value. Never invent an interview, a quote or a respondent count.
  4. Triangulate the two to five decisive claims (triangulation): two independent approaches, an independence check, an explanation of any gap. Revealed behaviour beats stated intent; the client's own estimate is a hypothesis, not a source.
  5. Outside view for every forecast: what comparable companies or projects actually achieved, or a statement that no comparable was found and a wider range.
  6. Counter-evidence: search for the strongest evidence against and record it.
  7. Verdict: confirmed, killed, reshaped (write the new hypothesis) or open (name what would settle it); evidence strength strong, medium or weak; one so-what sentence.
  8. Write research/<branch>.md:
# Research: <branch>, <date>
**Hypothesis:** | **Kill criterion:** | **Verdict:** confirmed / killed / reshaped / open; strength: strong / medium / weak
**So what:** <one sentence>

## Findings
| # | Claim (number, unit, scope, period) | Source (origin) | Date | URL or document | Type | Supports / contradicts | Strength |
## Triangulation of decisive claims
| Claim | Approach 1 | Approach 2 | Independent? | Final range | Strength |
## Outside view
| Forecast | Reference class | Range achieved | Where this case sits and why |
## Counter-evidence
## Assumptions and gaps
| Item | ASSUMPTION / unverified / missing | Range used | What would settle it | Owner |
## Data requests
| What | Format | Why it matters (which claim) | Fallback if missing | Owner |

In an engagement, update 02-hypotheses.md and add a line to 00-status.md.

Resume

Read every file in engagement/<client-slug>/, then report before doing anything else:

  1. The current answer and the status of every hypothesis.
  2. Inconsistencies between artefacts: ghost-deck pages relying on a killed hypothesis, workplan rows for dead branches, a status file behind the hypotheses file. Fix them or make them the first next step.
  3. Open partner gates. An open gate blocks the next phase; run it before more analysis.
  4. The next open workplan row with the highest impact, and what you need from the user.

Quality check before anything leaves

Run the checklist sections that fit (A and B for framing, C for research, A to D for answers) from partner-review-checklist and fix what fails. If partner-review is installed, recommend running it.

Rules

  • If the user mentions forwarded, attached or "the pack / thread / briefing" material, look for it in the working directory and the engagement folder before asking for it.
  • When you refuse part of a request (an invented number, a price from seller-only figures, a plug to hit a target), still deliver everything else the user asked for in the same reply.
  • No analysis before the problem statement is agreed (full engagement).
  • Every number has a source, date, unit and scope, or is labelled as an assumption. Precision follows evidence: give ranges.
  • A hypothesis without a number is not finished.
  • Never state an external benchmark, industry norm or market figure from memory as fact. Write what needs to be checked instead of a number.
  • Ask the user only what they can answer: facts, decisions, data. Do the structuring yourself.
  • Prefer a short, decided recommendation with stated uncertainty over a long, hedged one.
  • Keep the user informed in one line per step.

Ethics and limits

  • No confidential competitor information: no pretexting, leaked documents, or asking former employees or customers about non-public matters. No coordination of prices with competitors.
  • Respect privacy: no personal data beyond what the task requires.
  • Without web access you cannot verify external facts; say which claims need checking.
  • Legal, tax, audit, regulatory and technical due diligence need specialists; flag where they are needed.

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.