AM1010101/agent-skills

ICP

Help a solo founder turn a validated problem into a sharp Ideal Customer Profile: who exactly to serve, which segment to start with, the job they're hiring the product for, and who to disqualify.

How To Use This Skill

If the idea or problem isn't validated yet, use the discovery skill before this one — an ICP built on an unvalidated problem is fiction. Then identify the mode and read only that reference.

• Segment scoring: comparing possible customer groups and choosing a beachhead segment. Read segment-scoring.md.
• ICP builder: creating or refining a full Ideal Customer Profile — including Jobs to Be Done and qualification/disqualification rules. Read icp-builder.md.
• Output templates: producing a one-page ICP or GTM handoff. Read output-templates.md.

If the user is ready to decide what to build, how to position it, how to sell it, or how to price it, use the product skill after the ICP is clear. If it isn't yet clear what the founder needs, start from the founder router.

Working State

Used across several products with no shared memory, so persist to disk using the three-layer model in ../founder/CONVENTIONS.md §4.

• On start, ask which product, then read product/<product-slug>/state/discovery.md (the validated problem + evidence, from the discovery skill) and any existing state/icp.md, and build on them rather than starting over. If there's no validated problem yet, send the founder to discovery first. Scan log/ for due or contradicted predictions and offer to review them first.
• On finish, replace state/icp.md with the latest ICP — header it with Updated:, Confidence:, and Change triggers:. This is the file the product skill reads. When you make or revise a customer/segment call, also append a dated entry to log/decisions.md with a Prediction: and a Review: date (capture is your job, not the founder's).

Default Output Shape

When the user has not asked for a specific artifact, produce:

1. The recommended segment/ICP and its one-sentence definition.
2. The core job and the strongest fit evidence (plus biggest unknowns).
3. Who to pursue first (the ideal-of-the-ideal) and who to disqualify.
4. Confidence level and what would change the recommendation.