Profile Builder

Profile Builder

> Foundation skill of the discovery-phase skill pack · foundation group · writes discovery-context.md, the file every other skill in the pack reads.

Produces discovery-context.md — the single source of truth that every other skill in this pack reads. Without it, other skills will warn and offer to bootstrap inline (a 2–3 question mini-interview held in conversation, not written to disk). Running this skill first gives the best, most consistent results.

Activation rule

The moment this skill triggers, go straight to the interview. Do not summarise the skill, do not ask permission, do not explain the structure. The user already knows what they activated.

What this skill writes

A single file at the project root: discovery-context.md.

It contains 7 sections:

1. Client
2. Product / Initiative
3. Engagement
4. Access & Data
5. Expected deliverable
6. Constraints
7. Open questions about the engagement itself

See ./template.md for the exact structure.

Procedure

Step 1 — Check for existing context

If discovery-context.md already exists in the project root:

• Show the user a 5-line summary of what's there.
• Ask: "Do you want to refresh (re-interview from scratch), update (only the section I name), or keep (just confirm and exit)?"
• Route accordingly. Default to "update" if unclear.

If the file doesn't exist, proceed to Step 2.

Step 2 — Interview, in 4 batches

Use AskUserQuestion. Group questions to minimize round-trips.

Batch A — Client

• Client name and short description (1 sentence)
• Domain / industry
• Business model: B2B SaaS / B2C product / marketplace / services / internal tool / other
• Stage: pre-launch / early traction / scaling / mature / declining
• Decision-makers: who signs off on what (names/roles)
• Existing relationship: new logo / repeat client / staff aug / pre-sale prospect

Batch B — Product / Initiative

• What's being explored (one sentence)
• Current state of the thing: idea / spec / prototype / live but failing / live but pivoting
• Prior research / artifacts available (links, file names, or "none")
• Has the client already proposed a solution or feature list? (yes / no — and if yes, summarize it)

Batch C — Engagement

• Mode: pre-sale discovery sprint / paid discovery sprint / delivery-phase discovery / staff aug / internal request / other
• Timeline available: days / 1-2 weeks / 3-4 weeks / quarter / open-ended
• Runner role: solo BA / BA + PM / lead consultant / part of agency squad / internal product
• Budget context: free pre-sale / fixed-fee / time-and-materials / discovery-sprint fee / staff aug rate / unknown

Batch D — Access, Data, Deliverable, Constraints

• End-user access: direct / client-mediated / proxy-only / none
• Client SMEs available: list names + roles, or "none"
• Data we can request: analytics / support tickets / call recordings / past interviews / sales notes / NPS / other / none
• Tools BA has access to: list (Linear, Jira, Notion, Confluence, Mixpanel, etc.) or "TBD"
• Primary deliverable expected: proposal / SoW draft / discovery handoff / go-no-go memo / estimation / other
• Secondary deliverables (optional): list
• NDA / confidentiality status
• Languages, geographies, regulatory (GDPR / HIPAA / SOC2 / other / none)

Step 3 — Render the file

Render answers into the template. Keep it tight: each field one to three lines. No filler prose.

Add at the bottom:

## Open questions about the engagement itself
- (any answers the BA marked unknown / TBD become bullets here)

Step 4 — Confirm + exit

After writing, show the user a 6-bullet summary:

• Client: , ,
• Initiative:
• Mode: ,
• Access: ; SMEs: ; Data:
• Deliverable:
• Open questions logged:

Then state: "Discovery context written. Other skills in this pack will now run. Suggest starting with: problem-framing, stakeholder-mapping, or discovery-conductor if you want a guided sequence."

Anti-patterns

• Skipping the engagement section. "We'll figure out the engagement later" is how proposals end up unsigned. Force completion.
• Vague decision-makers. "Client leadership" ≠ name. Push for at least one named human per decision.
• Filler. If the BA doesn't know something, write "TBD — see open questions" and add to bottom list. Don't invent.
• Refreshing too eagerly. If the file exists, default to "update specific section", not "rewrite everything". Old context is hard-won.

What downstream skills consume

Every other skill in the pack reads discovery-context.md and runs the soft-gate pattern: if the file is missing, ask 1–3 skill-specific bootstrap questions inline, tag any unverified output field as [ASSUMED], and never block. See each skill's Step 1 for its specific bootstrap questions.

Specifically:

• problem-framing uses Client + Initiative + prior-solution sections
• stakeholder-mapping uses Client decision-makers + Engagement (two-tier map)
• research-planning, user-interviews, sme-workshops, secondary-research, support-data-analysis all branch on Access & Data
• estimation, proposal, sow-draft consume Engagement + Deliverable sections
• All deliverable skills read Constraints (NDA, regulatory) before writing

_log.md — pack-wide log schema

Every downstream skill appends one line to ./discovery/_log.md after writing its artifact, using a single canonical schema:

[ | YYYY-MM-DD] key1: value1; key2: value2; key3: value3

Rules:

• Brackets contain the skill name and the run date (ISO format, no time).
• Body is a flat list of key: value pairs separated by ;.
• Keys are skill-specific but should be short, lowercase, and stable across runs.
• One line per run. Multiple runs of the same skill produce multiple lines (newest at bottom).

This schema lets the conductor read _log.md to see what's been run, and lets deliverable skills (proposal, sow-draft, discovery-handoff, go-nogo-memo) recover the evidence chain without re-reading every artifact.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: stanislavnianko
• Source: stanislavnianko/product-discovery-claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.