product-discovery

Files (3)

• SKILL.md
• openai.yaml · agents/
• feature-discovery-session.md · references/

SKILL.md (first lines)

# Product Discovery

1. Identify the user, current task, observed friction and desired change in behavior.
   Reuse customer research, support evidence and constraints. Label facts, assumptions
   and unknowns; never invent interviews, demand, baselines or measured improvement.
2. Identify the uncertainty most likely to change the decision. Consider a process
   change, existing capability or smaller solution before committing to new software.
3. Choose the cheapest useful experiment or complete end-to-end slice. State the
   observable success signal, how it will be measured and what result would stop or
   change the approach. Label proposed targets; use real baselines when available.
4. Define scope and exclusions, the primary user journey, material error/recovery
   behavior and consumer-visible API needs. Separate usability, feasibility and demand
   questions; a working prototype establishes neither adoption nor business value.
   Name the product decision/acceptance owner and constraints from affected stakeholders.
5. Put acceptance scenarios and the next action at the project-designated planning
   home (for example, the assigned Jira ticket or GitHub issue). Link research/designs at their source. Ask only for information that
   changes a consequential decision; resolve routine reversible details independently.
6. Deliver a small increment, collect actual feedback and compare it with the success
   signal before expanding scope. Keep technical completion and observed impact separate.
   Propose priority from value, uncertainty, effort and dependencies; preserve conflicting
   feedback and the owner's actual decision rather than inventing consensus.

## Output

Return the problem, evidence/assumptions, chosen increment, exclusions, acceptance
scenarios, success signal and next decision in a compact form. Include alternatives
only when they explain a real tradeoff; create no extra roadmap or proposal by default.

## Optional example

[Stakeholder conversation](references/feature-discovery-session.md): use when interview
structure or acceptance extraction is the unresolved problem.

From atstaeff/ai-agents, MIT.

More from atstaeff/ai-agents