Icanpreneur

View saved version →

---
name: icanpreneur
description: 'Foundation for every task that uses the Icanpreneur tools: workspaces (called ideas), Canvas, customer segments, research, interviews, contacts, Buyer Personas, positioning, landing pages, sales decks, notes, exports and team access. Load it before the first Icanpreneur tool call in a conversation and before any billable or destructive call. Covers vocabulary, IDs, credits, waiting for generated work, retries, evidence rules and deletion.'
---

# Icanpreneur: how to use the tools

Icanpreneur helps founders, product teams, consultants and startup programs decide from customer evidence. Ground what you say in what is saved in the workspace, and be clear how strong that evidence is.

Load a workflow skill as well when the task matches:

| The user wants to                                                                           | Also load                                                                |
| ------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------ |
| A Buyer Persona's reaction to an offer, price, pitch or idea                                | [icanpreneur-persona-feedback](../icanpreneur-persona-feedback/SKILL.md) |
| Set up a workspace, segments, research or interviews; read findings                         | [icanpreneur-build-research](../icanpreneur-build-research/SKILL.md)     |
| Positioning, a landing page, a sales deck or a preview                                      | [icanpreneur-gtm-assets](../icanpreneur-gtm-assets/SKILL.md)             |
| Decide what to test or change; judge whether evidence supports something; update the Canvas | [icanpreneur-decide](../icanpreneur-decide/SKILL.md)                     |
| Review several startups together; prepare mentoring                                         | [icanpreneur-cohort-review](../icanpreneur-cohort-review/SKILL.md)       |

Notes, exports, team access and questions about saved work need only this skill. Use the user's words (workspace, Canvas, research, persona) and name a tool only when it helps. Some website terms map to the tools: IVA is Icanpreneur's assistant, "Validate customer problems" is `ProblemDiscoveryAndValidation` research, "Analyze won/lost deals" is `WinLoss`, "Understand buyer decisions" is `BuyerPersona`, "Talk with your persona" is `ask_buyer_persona`. Some website features have no tool: see [references/not-available.md](references/not-available.md).

## The rules that matter most

1. **Real versus simulated.** Synthetic interviews, synthetic contacts and persona answers are simulations. They give hypotheses, not validation, demand or traction. Always say how many sources are real, imported and synthetic.
2. **Show the source.** Every conclusion names its evidence: the insight or interview, whether it is real, imported or synthetic, and a count when the tool returns one. Leave missing quotes and counts missing. Quote only returned text, in the interview's language. Mark your interpretation as yours.
3. **Do what was asked.** Complete the workflow the user authorized and say which steps use credits. Ask only when a consequential choice is missing, when you would add scope they did not request, or when they asked to review first.
4. **Never guess.** Take IDs from results in this conversation. Do not invent names, quotes, numbers or proof. If a tool you need is unavailable, say so and do not claim the work was done.
5. **Credits.** Before the first billable call, check `get_ai_credit_balance`. At zero or below, do not start; explain that a positive AI-credit balance is required. Work already admitted can finish below zero. The billable tools are in [references/billing-and-async.md](references/billing-and-async.md); easy to miss are `validate_canvas_item`, `create_canvas_customer_segment`, `create_canvas_item` with omitted or empty `relatedItemIds`, and interview i

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 8 — file references are metadata, not package contents.

View saved version →

---
name: icanpreneur-build-research
description: 'Set up and run customer research in Icanpreneur (problem discovery and validation, win/loss, buyer persona, or other research): create a workspace from a business description, define or refine customer segments, plan interview scripts, create research, set up interviews with specific customers, import existing transcripts, add synthetic interviews, read findings, and turn interviews into a Buyer Persona. Use when the user wants to start research, add evidence, understand what customers say, prepare the next round, fix a misfit interviewee or split a segment. Load `icanpreneur` first. Do not use for general research advice that does not involve an Icanpreneur workspace, or for a simple lookup of saved research.'
---

# Build and run customer research

workspace → customer segment → interview script → research → interviews (real, imported, synthetic) → analysis

Analysis and insights are the common endpoint for every research type. Only `BuyerPersona` research has an automatically associated Buyer Persona; do not automatically expect, wait for or mention a persona for other types (an explicit request to compare findings with an existing persona is fine: read it with `get_buyer_persona`). Load [icanpreneur](../icanpreneur/SKILL.md) first; its rules apply to every step.

## 0. Scope

Do the whole piece the user asked for, say which steps use credits, and do not add billable steps beyond it.

| Scope                              | Stop when                                                                                                                                                                                |
| ---------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Set up                             | The research exists and is ready for interviews.                                                                                                                                         |
| Gather evidence                    | The requested interviews are created or imported and the user has what they need (links, counts).                                                                                        |
| Read results                       | The question is answered from saved analysis, with its evidence mix and `updatedOn`.                                                                                                     |
| To analysis (any type)             | `get_research_analysis` shows `Complete` and you have reported the findings. If still `Pending`, return the IDs and the next check.                                                      |
| To a persona (`BuyerPersona` only) | The persona you triggered is `Complete`, has usable content, and is shown with its evidence line. If it ends in `Error`, stop and say so. Do not present an older version as the result. |

Ask only when a choice is missing: segment, research type, language, script duration, or which interviews to add.

## 1. Workspace

`create_idea` needs `title`, `startingPoint` (`idea` or `existingBusiness`; ask if unclear), `languageTag`, and a `description`, `urls` or `fileKeys`. A logo is not context. Do not invent a description. If the user has no idea yet, finding one is not a tool; talk it through and create the workspace once there is a description.

- Call it once; never again to check. `create_idea` creates the initial Canvas during the call. After success, read `get_canvas` immediately; no separate readiness field or Canvas poll is needed. Logo generation may still run. A timeout or error can leave partial saved work; inspect before any retry. The Canvas is a first draft of hypotheses: show the segments and problems and invite corrections.
- A distinct growth opportunity may suit its own workspace so t

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 3 — file references are metadata, not package contents.

View saved version →

---
name: icanpreneur-cohort-review
description: "Review several Icanpreneur startup workspaces together for an accelerator, incubator, university program, mentor or client portfolio: a cohort overview of each team's customer, problem, value proposition and assumptions, evidence strength per team, common challenges, workshop topics, who needs individual mentoring, and a mentoring brief per startup. Use when the user names several startups, teams or workspaces and wants an overview, program review or mentoring preparation. Load `icanpreneur` first. Do not use for a single workspace (`icanpreneur-decide`) or for setting up research."
---

# Cohort review and mentoring briefs

Load [icanpreneur](../icanpreneur/SKILL.md) first; its rules and its multi-workspace reference ([several workspaces](../icanpreneur/references/multiple-workspaces.md)) apply. Reviews are reads: no credits, and nothing is changed unless the user asks for a note. For one workspace use `icanpreneur-decide`; its assumption tests apply to every team here.

## 1. Resolve the cohort

Find each named workspace with `list_ideas` (add `includeArchived` if needed). Ask if a name matches several. List any workspace you cannot open as inaccessible and carry on. If the user did not name the teams, ask; do not guess a cohort from every workspace they can see. Decide the question: **Canvas only** (target customer, problem, value proposition, assumptions) or **Canvas and research**. A Canvas-only request is complete from the Canvases: do not fetch research or personas.

## 2. Read every team the same way

For each team: `get_canvas`; if research is in scope, `list_research`, then per research its type, `get_research_analysis` (status, `updatedOn`), key insights and the evidence mix of completed interviews (real, imported, synthetic), de-duplicated by interview ID when projects share interviews; persona status if the question concerns it; `list_notes` for earlier reviews. Use the same criteria for every team, state them, and label every finding with its startup.

## 3. Assess each team

- **Customer and problem**: apply the platform tests; say if the segment is only a role or the problem is vague.
- **Assumptions**: Canvas content is a hypothesis. **Supported** only when real or imported interviews say it, with counts where established and interviews named where available. Synthetic-only is "simulated".
- **Evidence**: report interview stage and analysis state separately.
- **Biggest untested assumption**: the one that matters most if wrong, upstream first.
- **Next milestone**: one concrete step that closes it.

Interview counts do not measure startup quality. Real and supplied imported interviews can support customer findings; synthetic interviews remain hypotheses regardless of quantity. Do not rank teams on a single score or invent one. If asked for a priority order (for example where mentoring helps most), state the criteria and the basis for each placement.

## 4. Output

- **Cohort overview**: one row per startup (target customer, problem, value proposition, key assumptions). When research is in scope add the evidence mix, which streams exist (problem, buyer, win/loss) with project counts and `lastActivity`, and what is supported. Then common challenges, workshop topics and where mentoring helps most, each with the startups behind it and the reason.
- **Program review**: for each finding, the startup it comes from, whether it is supported, only simulated or needs testing, and what each team should focus on next.
- **Mentoring brief** per startup (**read [references/mentoring-brief.md](references/mentoring-brief.md) before writing one**): strongest evidence, biggest untested assumption, three questions for the mentor, next milestone.

List inaccessible workspaces at the top.

## 5. Rules

- Never present one startup's evidence as another's, or carry a finding between teams. If the output goes to the teams themselves, ask whether to 

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: icanpreneur-decide
description: 'Turn saved Icanpreneur Canvas, research and insights into a reasoned decision: what to test or change next, whether to prioritize a customer segment, whether positioning needs to change, whether the evidence supports a feature or claim, or whether more evidence is needed. Also updates the Canvas after new evidence and checks a pitch, deck or set of claims against the research. Use when the user asks what to do next, which assumptions are riskiest, whether the evidence supports something, or for a decision brief or recommendation. Load `icanpreneur` first. Do not use to run or set up research (`icanpreneur-build-research`), to ask a Buyer Persona a question (`icanpreneur-persona-feedback`), or to generate assets (`icanpreneur-gtm-assets`).'
---

# From evidence to a decision

Research ends with findings. This skill turns them into a choice the user can explain and defend: what the evidence supports, what it contradicts, what is untested, and what to do next. Load [icanpreneur](../icanpreneur/SKILL.md) first; its rules apply. **Before writing the brief, updating the Canvas or checking a pitch, read [references/decision-brief.md](references/decision-brief.md)** (template, criteria, procedures).

By default this skill only reads. Write only when asked: edit Canvas items, or save a note. Do not start research, regenerate assets or ask the persona unasked; offer them as next steps. For several workspaces use `icanpreneur-cohort-review`.

## 1. Pin down the decision

State the decision in one line and the segment it concerns ("Prioritize segment A?", "Does the evidence support our positioning?", "What should we test next?"). If either is unclear, ask one short question. Find the workspace.

## 2. Read what exists (free)

1. `get_canvas`: the segments, problems and other items. These are the assumptions.
2. `list_research` and `get_research` for the segment.
3. `get_research_analysis` (status, `updatedOn`), `list_research_insights` (all sources/quotes, unpaged) and `get_research_quality`.
4. The evidence mix of each research you use: paged to `totalCount`, de-duplicated by interview ID when projects share interviews.
5. `list_notes` for earlier decisions on the same question.
6. `get_buyer_persona` status, if a persona exists.

If analysis is `Pending` or older than the newest interviews, say the findings may not include them. Report interview stage and analysis state separately.

## 3. Sort and judge

Keep five kinds apart: real or imported interviews ("customers said…"), synthetic interviews (hypothesis), Canvas items (assumptions), persona answers (simulated, never demand), and your interpretation.

Give each assumption a status and its basis:

- **Supported**: real or imported interviews say it.
- **Contradicted**, **mixed**, **only simulated**, or **untested**.

Then:

- Cite the evidence available and counts where established. Leave missing quotes and counts missing.
- Do not score evidence, invent a threshold or say "validated". Say when a finding rests on one interview or only synthetic ones.
- Check the segment and problem against the platform definitions first. A segment that is only a job title, or a problem with no stated alternative, is itself a weakness.
- Look across all research types, and match each assumption to the stream best placed to test it (reference).

## 4. Prioritize

Rank by impact if wrong and by uncertainty, with reasons in words. Segment and problem come first when they are weak or the decision is about direction. For a price, feature or message decision, prioritize the assumptions behind that decision. Do not send the user back to segment research when the question is not about the segment.

## 5. Recommend

Lead with the actual recommendation in the user's terms ("prioritize segment A", "refine pricing", "defer the feature", "collect more evidence first"), then the reasons. Describe the evidence as **strengthenin

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: icanpreneur-gtm-assets
description: 'Turn a Buyer Persona into go-to-market assets in Icanpreneur: positioning and messaging, a landing page, a sales deck, and a webpage or Google Slides preview. Use when the user wants positioning, messaging, landing-page copy, a sales or pitch deck, wants to edit or refine those, regenerate them, or get a preview link. Load `icanpreneur` first. Do not use for generic copywriting, building a custom website, or an unrelated presentation. Covers the generation order, which ID goes where, editing without regenerating, previews, and what the assets can and cannot claim.'
---

# Positioning, landing page, sales deck and previews

Load [icanpreneur](../icanpreneur/SKILL.md) first; its rules apply.

```
Buyer Persona → Positioning and messaging → Landing page ─┐
                                          └→ Sales deck ──┴→ Preview (webpage / Google Slides)
```

- Positioning needs a persona (`generate_positioning_and_messaging`, with `buyerPersonaId`).
- A landing page or deck needs a **positioning document** (`generate_landing_page`, `generate_sales_deck`, with `positioningDocumentId`).
- A preview needs a landing page or deck (`generate_buyer_persona_document_preview`, with `buyerPersonaDocumentId`). Positioning has no preview.

Every document is one shared resource. Its `id` is the `buyerPersonaDocumentId` for read, update, delete and preview, and the `positioningDocumentId` when it is a `PositioningAndMessaging` document. It is not the content `documentId`. Never pass a persona ID to the page or deck generators.

Done means the documents asked for are `ready` with `error: false`, requested previews are `ready` with `error: false` and a link, and the user knows what evidence the assets rest on. Generate only what was asked: "positioning" is not a request for a page and a deck.

## 1. Check the starting point

1. Find the workspace and persona. No persona → `icanpreneur-build-research`.
2. Positioning generation requires `status: Complete` and a nonempty `description` (`get_buyer_persona`). This applies to existing personas too: Pending/Error is rejected even when fields from an older profile remain. Wait briefly or report the block; there is no tool to select a previous completed version. An empty Complete persona is not ready, and Complete does not prove recent interviews are included.
3. Count the real and synthetic interviews in the source research (`icanpreneur` → `references/evidence.md`). If mostly synthetic, say in the hand-over that the assets express a hypothesis. This is disclosure, not an approval step.
4. `list_buyer_persona_documents` shows what exists. Each generation makes a new document; if it is unclear whether to reuse the existing positioning, ask.
5. Use the workspace or research language unless asked.

## 2. Generate

These calls are billable. Say which steps use credits, then run the chain the user asked for without stopping at each step.

1. Positioning; wait for `ready: true` and `error: false` (`get_buyer_persona_document`; `text` can be null while saved content is unavailable). On `error: true`, report it. Never write the positioning, page or deck yourself in place of the generator.
2. Pause to show the positioning only if the user asked to review or refine it first. Otherwise continue. Pages and decks capture their source when generated, so the time to change the positioning is before they are built.
3. Landing page and/or deck from the positioning's `id`. Wait for `ready`.
4. A request for a preview is the authorization. Do not ask again because it is billable.

**Before editing any generated text, or generating or refreshing a preview, read [references/editing-and-previews.md](references/editing-and-previews.md).** Prefer editing the Markdown over regenerating.

## 3. Review after new evidence

When new interviews have changed the persona, or the user asks whether the assets still fit, compare before regenerating. Ch

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: icanpreneur-persona-feedback
description: "Get a Buyer Persona's reaction to an offer, price, pitch, message, landing-page copy, feature or idea using the user's saved Icanpreneur research. Use when the user wants to test, pressure-test or get feedback from their buyer persona, compare options, anticipate objections, or improve material using the persona's reaction. Load `icanpreneur` first. Do not use to build a persona (`icanpreneur-build-research`), to generate positioning or pages (`icanpreneur-gtm-assets`), or to answer as a persona yourself."
---

# Ask your Buyer Persona

Put the user's work in front of a persona built from their research and get a reasoned reaction. The persona service writes the answer. Your job is to ask a good question, say how much weight the answer deserves, and help the user act on it. Load [icanpreneur](../icanpreneur/SKILL.md) first; its rules apply.

Done means the user has the persona's reaction, your reading of it, what to test with real customers, and a saved note if they want one. If there is no persona, go to `icanpreneur-build-research`.

## 1. Find the persona

1. Find the workspace (`list_ideas`), then `list_buyer_personas`. Each entry shows name, title, source research, segment and `status`.
2. One persona: use it and say which. Several: ask which, unless the user named one. Offer to ask each if more than one fits.
3. Read `get_buyer_persona`: Complete must also have populated profile content. Pending/Error may retain older or incomplete fields, not a guaranteed selectable historical snapshot. For the build you just triggered, wait for Complete and usable content; stop on Error. For an existing populated profile, disclose its state before an authorized question and stop if it fails. No usable persona: it comes from BuyerPersona research and analyzed interviews.

## 2. Know what it rests on

Once per persona per conversation (free): count the completed interviews in the persona's source research by `synthetic` and `imported`, paged to `totalCount` (procedure: `icanpreneur` → `references/evidence.md`). Keep one line: "The source research currently contains 8 completed interviews: 3 real (1 imported) and 5 synthetic." This is the research now, not proof of what the answering version was built from, so never say "this persona is based on…". A persona answer has no sources field: never invent citations. If everything is synthetic, say so before asking; the answer is a simulated view.

## 3. Ask

- Put the user's supplied text verbatim inside `question`. Include source URLs in `question`; the persona service may retrieve those public webpages. Uploaded attachments use `fileKeys` with usage ChatFile and actual bytes. Do not invent an attachment from a summary.
- **Before writing the question, read [references/question-patterns.md](references/question-patterns.md).** Ask open, specific questions about reaction, hesitation, missing information and what would convince them. No leading questions, no requests for numbers or shares, no "pretend you are".
- Check the credit balance before the first question (each is billable).
- Call `ask_buyer_persona`. Keep the returned `conversationId` for follow-ups. If `usesLatestPersona` is false, say the conversation started at an earlier persona revision and offer a new conversation. This does not establish an immutable historical evidence snapshot. A lost ID means starting a new one.
- To compare options, put them in one question labelled A, B, C (one charge). If the choice matters, offer a second run with the order changed.
- To test a different idea without the earlier conversation influencing it, start a new conversation.
- After an unknown outcome, do not ask again. Tell the user it may have gone through and ask whether to retry. On `INSUFFICIENT_AI_CREDITS`, explain that a positive AI-credit balance is required and stop.
- If the call fails, never answer as the persona yourself.

## 4. Present the answer

Ke

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 1 — file references are metadata, not package contents.