A power coach, performance coach, and business therapist in one installable skill — with memory, modes, and a sourced evidence base.
Generic AI is agreeable and amnesiac. Ask it about your business and you get encouragement, seven suggestions, and no follow-up — ever. Founders don't need another pleasant chat; they need direct challenge grounded in evidence, ONE commitment at a time, and someone who remembers what they promised last week and asks about it first. That is a process, and process can be specified and installed. This repository is that specification.
| Without | With |
|---|---|
| "Great progress! Here are 7 ideas…" | The single constraint gating revenue, named, with the founder who faced it and what they did. |
| Every session starts from zero | Session starts by auditing last session's commitment: kept, broken, or renegotiated. |
| Invented statistics delivered confidently | Every claim carries an evidence tier — [peer-reviewed] / [practitioner] / [inference] — and benchmarks name their vertical. |
| Vague intentions ("I'll try to launch soon") | One commitment, as an if-then plan, with a deadline, written to a ledger. |
| Endless agreeable conversation | A session with no number, no hard thing, and no commitment doesn't close — drift itself gets named. |
The skill routes each session by what the founder says:
| Mode | Triggered by | What happens |
|---|---|---|
| Intake | First run | Discovery interview → builds the founder-context file, sets the directness contract (how hard to push, 1–10), ends with a real commitment |
| Quick Counsel | One tactical question | Direct answer with the doctrine citation — no ceremony |
| Validate | Pre-revenue, "is this idea worth building" | Evidence-ladder placement, Mom-Test discovery, proceed/pivot/kill gates set before the test |
| Strategy Session | "What should I focus on" (default) | Bill Campbell's five-topic 1:1: personal check, gate metrics, the one constraint, doctrine direction, close |
| Unblock | "I'm stuck / overwhelmed / can't start" | Names the block, administers the matched intervention (WOOP, fear-setting, if-then, complicity question) step by step |
| Decision Gate | "Should I X or Y" | Reversibility triage, compressed fear-setting, doctrine check, a clear recommendation — recorded and audited later |
| Weekly Review | "Weekly review / check in" | Ledger audit, week-over-week gate numbers, tripwire check, focus accounting, boundary check, top-3 → one commitment |
| Foresight | "What could go wrong", a new quarter or launch ahead | Rolling 30/60/90 premortem → tripwires with pre-agreed responses → blameless postmortem when one fires |
| Life | Burnout, relationship strain, "who am I without this" | The founder as a person: values, relationships, money psychology, sleep, identity — evidence-tiered, referral triggers armed |
| Crisis | Cash crunch, platform ban, defect, legal notice | Strict triage: protect cash → size exposure in numbers → communicate → fix → build the system the crisis teaches |
| Hard Truth | "Audit me / give it to me straight" | Full audit against the doctrine laws, failure patterns, and the founder's own recorded tendencies |
SKILL.md— the always-loaded constitution: voice rules, mode routing, session close protocol.references/— read-only knowledge modules loaded per mode: the 12 doctrine laws, 12 founder dossiers, stage playbooks ($0→$1M→$10M→$100M), unblocking protocols, seven entrepreneur modules (validation, finance, marketing, sales & pricing, hiring, crisis, operating cadence), and seven research-driven expansions (B2B playbook, cross-model benchmarks, failure taxonomy, blocker map, foresight protocols, life coaching, discipline & execution).state/— the memory that makes candor personal: founder context, a commitment ledger audited every session, and an append-only metrics history. Created from templates at Intake, gitignored, never committed.
Eleven deep-research reports (450+ cited sources) sit in references/research/: documented founder case studies (Gymshark, Ridge, Native, Spanx, RXBAR, ButcherBox, Obvi, AG1, True Classic, MVMT, Liquid Death, HexClad), stage benchmarks for bootstrapped DTC, peer-reviewed coaching/performance psychology (implementation intentions, WOOP, feedforward, radical candor), and B2B founder case studies (Mailchimp, Basecamp, Ahrefs, ConvertKit, Atlassian, Zapier, DesignJoy, McKinsey/Bower, plus manufacturing and wholesale cases). The distilled modules cite them; the mentor labels evidence tiers out loud and names the vertical a benchmark came from before applying it to yours.
Claude Code — personal (all projects):
git clone https://github.com/makesupply/mentor.git ~/.claude/skills/mentorWindows: clone into %USERPROFILE%\.claude\skills\mentor\.
Claude Code — per project: clone or copy the folder to <repo>/.claude/skills/mentor/ instead.
claude.ai: zip the folder so SKILL.md sits at the top level of the zip, then upload under Settings → Capabilities → Skills. Note: on claude.ai the ledger can't persist as a file between chats — keep it in a Project's knowledge instead; Claude Code is the full experience.
First session: invoke /mentor. It will notice there's no founder context and run Intake — a discovery interview that builds your files and ends with your first commitment. From session two onward, expect the ledger audit first.
- Not therapy or clinical care. The skill carries a wellbeing floor: persistent clinical-grade strain gets a plain "get real professional support," then continued business help.
- Not financial, legal, or tax advice. Consequential irreversible moves in those domains always get "professional check" attached to the recommendation.
- Not timeless. Research gathered August 2026; benchmarks and revenue figures age. The mentor says "as of the research" and verifies live when precision matters.
- Not universal doctrine. The evidence base is deepest in bootstrapped DTC e-commerce. The laws carry a scope note; the mentor translates rather than transplants for other business models, and says so.
Edit state/founder-context.md freely — it's yours, including the directness calibration (1–10). Adjust or extend references/ modules for your industry; add modes by extending the routing table in SKILL.md. Keep user data out of references/ — the state/content split is what makes updates safe.
SKILL.md # constitution: voice, modes, close protocol
references/
doctrine.md # the 12 laws, with scope note
founder-dossiers.md # 12 documented founders, indexed by situation
stage-playbooks.md # $0→$1M→$10M→$100M gates and levers (DTC)
stage-playbooks-by-model.md # SaaS, agency, marketplace, creator, local benchmarks
b2b-playbook.md # B2B structure diagnosis, patterns, dossiers
failure-taxonomy.md # real base rates, failure causes, early-warning signals
foresight-protocols.md # premortem, tripwires, reference-class checks, postmortems
life-coaching.md # whole-person coaching: values, relationships, money, sleep, identity
discipline-and-execution.md # habit engineering, accountability, environment design, learning science
unblocking-protocols.md # block table + 13 administered interventions
blocker-map.md # differential diagnosis of 10 blockers, replication-honest
validation-playbook.md # idea → first paying customer
finance-fundamentals.md # unit economics, cash, funding ladder
marketing-acquisition.md # channel doctrine, creator playbook, paid discipline
sales-and-pricing.md # pricing architecture, raise protocol, negotiation
hiring-and-delegation.md # delegation ladder, leverage menu, keeper test
crisis-protocols.md # triage order + per-crisis table
operating-cadence.md # daily/weekly/quarterly/annual rhythm
research/ # the eleven sourced research reports (450+ citations)
state/ # your data: context, ledger, metrics (templates + README)
examples/ # sample sessions showing the register
SemVer: MAJOR = mode-system/voice/state-format changes; MINOR = new modules, modes, or research refreshes; PATCH = corrections. Changelog in Keep a Changelog style. Dated research claims are never silently edited — refreshes bump MINOR and update the research-date line.
MIT-0 — use, modify, redistribute freely. Attribution appreciated, not required.