specialized-profile-builder

name	specialized-profile-builder
description	Builds a Specialized Memory Profile (Simmons Class 2 architecture) for one operator-domain pair so future AI sessions reason from inside the operator's actual values, thinking style, blindspots, frameworks, decision history, constraints, and goals — not generic defaults. Two MVP branches — Strategic/Decision-Making and Learning/Coaching/Creative — selected from the prototypical problem. Uses Disconfirming Evidence Questioning to surface blindspots the operator can't self-report. Produces 4 artefacts (Profile Card, Usage Guide, Prompt Templates, Refresh SOP) saved to 00 - COMMAND CENTER/Foundational Docs/Profiles/. USE THIS SKILL when user says 'build a profile', 'specialized memory profile', 'operator profile', 'profile for my [domain]', 'the AI doesn't know how I think', 'I keep re-explaining my context', 'make AI more domain-specific', or wants Claude domain-aware for repeated sessions.
allowed-tools	Read, Write, Edit, Glob, Bash, AskUserQuestion

Specialized Profile Builder

Workflow: Wizard + Feedback Loop. Four-input intake → branch selection → dimension-weighted questioning with Disconfirming Evidence pushback → synthesis → four output artefacts saved to the vault. One question (or tight cluster) at a time; draft options before locking; iterate until the operator says it lands.

This is a process skill — the output is a 7-dimensional self-model, not a deterministic transform. Quality comes from the quality of the conversation, not from rigid templates.

---

Why this skill exists

Generic AI prompting plateaus once the operator has the basics. The next leverage layer (Simmons Class 2: Specialized Memory) is domain-specific operator profiles — pre-loaded context that lets future AI sessions reason from inside the operator's actual values, frameworks, decision history, constraints, and goals rather than from the model's defaults.

The bottleneck is not more prompting skill. It is the AI not knowing who it's prompting for. Profiles close that gap.

The skill produces operator self-models, not project documents. Project docs (CCC ICP, EI offer, LONNEL positioning) live in gtm-discovery outputs. Profiles cite those docs as inputs but answer a different question: how does this operator actually think and decide in this domain?

---

What it produces

For each invocation: ONE specialized profile for ONE (operator × domain) pair. Outputs land in 00 - COMMAND CENTER/Foundational Docs/Profiles/[Domain Name]/ as four files:

1. Profile Card.md — the 7 dimensions, 2-5 sentences each, dated and versioned. System-prompt-embeddable — designed to be loaded as context into future sessions.
2. Usage Guide.md — when and how to invoke this profile; example prompts; integration notes.
3. Prompt Templates.md — 3-5 ready-to-use prompts that pre-load the profile (e.g., "Using my CCC strategic profile, respond to this client objection: …").
4. Refresh SOP.md — when to refresh the profile, how (full re-run vs. dimension-specific update), what triggers an update.

---

Phase 0 — Intake (4 required inputs before questioning starts)

Use AskUserQuestion to collect these efficiently. Group as one tool call where the schema permits, otherwise sequential. Don't proceed to Phase 1 without all four.

1. Domain name — free text. Examples: "CCC strategic decision-making", "Leadership of LONNEL business with Jack and Anitta", "Personal learning of new technology", "Thought partnership for content ideas". Push for specificity if the operator says "business" or "work" — what subset of business?

2. Prototypical problem — free text, 1-3 sentences. A real, current, concrete decision or challenge in this domain. This grounds every subsequent question against a lived example, not abstractions. If the operator can't name one, that itself is signal — push: "What's one decision in this domain you've made (or avoided making) in the last 30 days?"

3. Use case — single-select pills:

   • Personal — profile is for the operator's own future sessions
   • Team — profile represents how the operator + their team approaches this domain (multi-stakeholder)
   • Product — profile is for a product/service Claude will use (e.g., a CCC client)

4. Refresh cadence — single-select pills (defaults to Monthly):

   • Weekly — high-volatility domain, new evidence each week
   • Monthly — most common; balanced freshness vs. effort
   • Quarterly — slow-moving domain, infrequent updates
   • Ad-hoc — refresh only when a decision surprises the operator

Save the intake answers as a temporary scratch file at /tmp/profile-intake-<domain-slug>.json for use in later phases.

---

Phase 1 — Branch selection

Based on the prototypical problem, infer which branch fits. Two branches in MVP:

Branch	When it fits	Front-weighted dimensions
Strategic / Decision-Making	The prototypical problem is about a decision, trade-off, allocation, plan, or judgement call (CCC engagements, LONNEL strategy, GTM choices, hiring, pricing, scope decisions)	Decision History · Frameworks · Constraints
Learning / Coaching / Creative	The prototypical problem is about understanding something new, building skill, reflecting on patterns, or producing creative output (learning a new technology, coaching a relationship, content ideation, identity work)	Values · Thinking Style · Blindspots

Important: "Front-weighted" means more questions land on those dimensions, not that the others are skipped. All 7 dimensions get covered in the final Profile Card.

If the prototypical problem genuinely straddles both (e.g., "How do I decide what to learn next?"), default to Strategic but ask the operator at branch-selection time: "This sits between Strategic decision-making and Learning/Creative work. Which lens feels more accurate for the kind of profile you want?"

If a third branch is clearly needed (e.g., a Leadership/Relational profile that doesn't fit either), surface that to the operator and skip to v0.2 territory — don't force-fit MVP branches.

For deeper branch-specific questioning sequences, read references/branches.md.

---

Phase 2 — Dimension-weighted questioning

For each of the 7 dimensions, ask 1-3 questions calibrated to the branch. Apply Disconfirming Evidence Questioning at each dimension — push the operator where their answer feels too clean.

The 7 dimensions, in canonical order:

#	Dimension	Question shape (general)
1	Values	What's non-negotiable in this domain? What would you walk away from a deal/decision/relationship over?
2	Thinking Style	How do you process information here — fast/intuitive, slow/deliberate, written/spoken, alone/with input? When do you trust your gut vs. defer to analysis?
3	Blindspots	Where do you know you systematically miss or distort? Where have you been corrected most often? What do trusted advisors push back on?
4	Frameworks	Which mental models do you reach for in this domain? Which do you trust most in conflict situations?
5	Decision History	What's the most consequential decision you've made in this domain? What's the pattern across your last 3-5 decisions here?
6	Constraints	What can't you spend (time / money / attention / risk)? What are the non-negotiable obligations?
7	Goals	What does "success" look like in this domain over the next 6-12 months? How will you know you got there?

Disconfirming Evidence Questioning (the signature move)

After every initial answer, push where it feels too clean. The operator will give a "should-be" answer first; the true answer surfaces under pressure. Use one of these prompts:

• "That's a clear value. When have you violated it? What did that situation tell you about how you actually decide vs. how you think you decide?"
• "You said you reach for [framework] — when did it fail you? What did you learn from that failure?"
• "You said your blindspot is X. Who's told you that? What's the evidence? Or is X just what you've named to look self-aware about?"
• "You said the constraint is time. If I gave you 10× the time tomorrow, what would actually change? If nothing — time isn't the real constraint."

This is what surfaces real Blindspots specifically (the operator can't self-report what they don't see) and refines every other dimension.

Branch-specific question depth

• Strategic branch: spend 50%+ of questioning time on dimensions 4-6 (Frameworks · Decision History · Constraints). Decision History is where the pattern of past choices reveals priors better than any self-report.
• Learning/Creative branch: spend 50%+ of questioning time on dimensions 1-3 (Values · Thinking Style · Blindspots). The operator is figuring out who they are in the domain; the work is meta-cognitive.

For full branch-specific question banks, read references/branches.md.

Pacing

One question (or tight 2-3-question cluster on the same dimension) per turn. Wait for the operator's answer. Apply Disconfirming Evidence pushback. Move on once the answer feels earned, not generic.

Refuse to accept generic answers. "I value honesty" is not a value — every operator says that. "I close the laptop and walk away when a client lies about scope" is a value. Push until the answer is operator-specific and example-grounded.

---

Phase 3 — Synthesis

When all 7 dimensions have earned answers (typically 25-40 minutes of conversation):

1. Read every existing artefact the operator named during questioning that could ground a dimension. If they cited the CCC ICP doc — read it. If they cited a [[Decisions & Rules]] entry — read it. If they cited a thinker (Simmons, Stutz, Tegmark) — note the lineage. This makes the profile cite-aware, not just self-reported.

2. Draft the Profile Card (see Output Formats below). Each dimension: 2-5 sentences, operator-specific, example-grounded.

3. Show the draft to the operator before saving. Critical step. Ask: "Here's the draft Profile Card. Read it as if a future Claude session were loading it. What lands? What's wrong? What's missing?" Iterate until the operator says it captures them.

4. Once locked, generate the other three artefacts (Usage Guide, Prompt Templates, Refresh SOP) — these are derived from the Profile Card and the intake, less iteration needed.

For full output templates, read references/templates.md.

---

Phase 4 — Save & integrate

1. Folder: 00 - COMMAND CENTER/Foundational Docs/Profiles/[Domain Name Slug]/. Slug the domain name (replace spaces with -, lowercase).

2. Files:

   • Profile Card.md
   • Usage Guide.md
   • Prompt Templates.md
   • Refresh SOP.md
   • _intake.md (the original 4 intake answers + branch decision, for audit trail)

3. Frontmatter on Profile Card.md:

   ---
   type: specialized-memory-profile
   operator: "[[Daniel Förster]]"  # or the named operator
   domain: "[domain name]"
   branch: strategic | learning-creative
   use_case: personal | team | product
   refresh_cadence: weekly | monthly | quarterly | ad-hoc
   created: YYYY-MM-DD
   last_refreshed: YYYY-MM-DD
   version: v0.1
   related:
     - "[[Three Brains Charter]]"
     - "[[The Five Stages of Third Brain Mastery — Curriculum]]"
     # plus any domain-specific docs the profile cites
   tags: [specialized-profile, simmons, second-brain]
   ---
   

4. Cross-link from related vault docs. If the profile is for a CCC strategic domain → wikilink from CCC foundations. If it's a leadership-of-LONNEL profile → wikilink from LONNEL person files. The profile shouldn't be an orphan note.

5. Update _index.md in the Profiles folder (create if missing) listing all profiles with their domain + branch + last_refreshed date.

6. Confirm the save with a Cowork file link (computer://...) so the operator can open the Profile Card and verify.

---

Integration with other skills (post-build, no action needed during build)

Once the Profile Card exists, these skills should detect and use it:

• second-brain-connect — when running cognitive operations in this domain, load the Profile Card as system context to ground operations in the operator's actual priors. Add detection: glob for matching profile in 00 - COMMAND CENTER/Foundational Docs/Profiles/ based on the target domain.
• third-brain-mastery — Stage 2 (Perspective Prompting) of the Five Stages curriculum requires specialized profiles. After this skill completes, surface to the operator: "You've now built one of the profiles required for Stage 2. Stage 2 unlocks once you have 3 profiles."
• ccc-mental-model-recipes — when the operator's preferred Recipes are noted in the Frameworks dimension, auto-suggest those when matching task shape.
• ccc-sales-prep / ccc-discovery-roleplay — a CCC sales profile sharpens these skills' outputs. When run for a CCC engagement, check for a CCC strategic profile and load it.
• gtm-discovery — parallel, not merged. If the operator's domain has a GTM venture (CCC, EI, LONNEL), suggest running gtm-discovery for the venture's market-facing docs separately. The profile cites those docs; they answer different questions.

---

Output Formats

Profile Card.md template

# Specialized Memory Profile — [Domain Name]

**Operator:** [[Operator Name]]
**Domain:** [domain name]
**Branch:** Strategic / Learning-Creative
**Version:** v0.1 (created YYYY-MM-DD)
**Last refreshed:** YYYY-MM-DD

> *Load this Profile Card as system-prompt context before working with the operator in this domain. Future Claude sessions: cite specific dimensions when offering analysis, recommendations, or pushback.*

---

## 1. Values
[2-5 sentences. Operator-specific, example-grounded. Front-weight in Learning/Creative branch.]

## 2. Thinking Style
[2-5 sentences. How they process information in THIS domain — not generic.]

## 3. Blindspots
[2-5 sentences. Surfaced via Disconfirming Evidence Questioning. NOT just "I'm too detail-oriented".]

## 4. Frameworks
[2-5 sentences. Mental models reached for in THIS domain. Cite vault wikilinks where they exist.]

## 5. Decision History
[2-5 sentences. Pattern across last 3-5 consequential decisions. Includes specific examples.]

## 6. Constraints
[2-5 sentences. Real constraints — time/money/attention/risk/obligations — that pre-shape choices.]

## 7. Goals
[2-5 sentences. Success criteria over 6-12 months. How the operator will know they got there.]

---

## How to use this profile

[1-2 sentences pointing to Usage Guide.]

*Refresh cadence: [weekly|monthly|quarterly|ad-hoc]. See Refresh SOP.*

Usage Guide template, Prompt Templates structure, Refresh SOP

For the full templates of the other three output artefacts, read references/templates.md.

---

Rules (Update This Section When Things Go Wrong)

1. Never produce a Profile Card without showing the draft to the operator first. A profile that the operator hasn't validated will not be loaded into future sessions — it's dead artefact. Iteration in Phase 3 is non-negotiable.

2. Never accept a generic answer to a dimension question. "I value integrity" is not a Value. "I walked away from a 50K engagement when the prospect lied about scope" is a Value. Push until specific.

3. Never skip Disconfirming Evidence Questioning. It feels uncomfortable to push back; that discomfort is what surfaces Blindspots. The skill's value is mostly here — without it, you produce a sanitised LinkedIn-bio version of the operator.

4. Never merge profile-building with project discovery. If the operator starts answering Values questions with their company's mission or their offer's positioning, gently redirect: "That's the venture's positioning. I want yours — the operator behind the venture. What do YOU value here, separately from what the venture sells?"

5. Never invent dimensions the operator didn't earn. If after Phase 2 a dimension is still thin (e.g., they couldn't name a Decision History pattern), say so on the Profile Card: "Dimension under-developed at v0.1; revisit at next refresh." Padding with generic content destroys trust in the artefact.

6. Never start questioning without all 4 intake inputs. No prototypical problem = no grounding for questions, and the profile becomes abstract. Stop and collect.

7. Never default to Personal when the operator means Team. A team profile (e.g., "How LONNEL leadership decides") needs who's in scope — Daniel + Jack + Anitta? Daniel only? Push for clarity.

8. Always cite vault wikilinks where they exist. If the operator names [[The Operator as Bottleneck]] as a framework they use, wikilink it in the Profile Card. The profile becomes part of the graph, not a standalone orphan.

9. Always save the intake to _intake.md in the profile folder. Future audits and v0.2 (domain-aware branching) need to know what triggered which branch.

10. Always offer multiple options at branch-selection ambiguity. If the prototypical problem straddles branches, name it explicitly and let the operator choose — don't silently pick.

---

Self-Improvement

When the operator corrects a draft Profile Card or rejects an output:

• Note which dimension was too generic, too specific, off-tone, or missing key context
• Add a Rule above if the failure mode could repeat
• Save the corrected version as a reference example in references/examples/ for future runs

When a Profile Card is approved on first draft:

• Note which framings of the questioning produced the strongest dimensions
• Save the conversation pattern as a reference example
• Surface to the operator whether the approved profile should become a template for similar domains (e.g., approved CCC strategic profile → template for "Strategic / Founder-led-business" use case)

When operators report 4+ weeks after building that the profile is being loaded into future sessions and producing better outputs:

• Capture the outcome in the Refresh SOP (what to preserve at refresh time)
• Use that signal to prioritise v0.2 features (which domain branches to add next)

When operators report the profile is not being loaded or is being ignored:

• That's a v0.3 signal — the closed-loop infrastructure (Decision Logger trigger + Refresh Orchestrator + Outcomes Tracker) is what closes this gap
• Capture the failure mode in references/examples/ and surface to the next v0.3 build session

This skill is never finished. The more profiles get built and used, the better the questioning sequences become. Treat the references/ folder as a living archive.

---

Anti-patterns table (the 3 most common failure modes)

Situation	Broken output	Root cause	Fix
Operator gives clean, sanitised answers throughout	Profile reads like a LinkedIn bio — generic, defensible, useless	No Disconfirming Evidence pushback applied	Re-run with explicit pushback at every dimension; expect 2-3 rounds before earning the real answer
Profile dimensions all 5+ sentences, very thorough	Profile is too long to load as system-prompt context (>800 tokens for the card alone)	Treating Profile Card as exhaustive documentation rather than embeddable context	Constrain to 2-5 sentences per dimension; move overflow into the Usage Guide
Operator says "I don't know" to several dimensions	Profile has multiple "Dimension under-developed" markers	Either the prototypical problem was too narrow OR the domain is too new for a profile	Either deepen via more time on dimensions that did land, OR defer the profile and recommend the operator try ad-hoc prompting in this domain for 4 weeks first, then re-build

---

When NOT to use this skill

• Project / venture documents — use gtm-discovery instead. ICP, positioning, voice, offer = market-facing artefacts, not operator profiles.
• Single-session context — if the operator is asking a one-off question, a Frontier Prompting move is faster than building a profile.
• Brand-new domains — if the operator has < 4 weeks of experience in a domain, the prototypical problem won't be grounded enough; defer the profile build.
• Pure execution checklists — if what's actually needed is a SOP, route to ccc-sop-creator.

---

Skill v0.1 created 2026-05-06 from Simmons AI Second Brain Course Class 2 framework-extract. Companion to [[Vector DB Pilot — Project Plan]] (semantic profile retrieval upgrade) and [[Specialized Profile Builder — Skill Placement Decision]] (architecture decisions). v0.2 adds domain branches; v0.3 adds closed-loop infrastructure (Decision Logger trigger, Refresh Orchestrator, Outcomes Tracker).