--- name: setup-ramp preamble-tier: 0 version: 1.0.0 description: | First-time setup for /ramp. Interviews the user to build their personal context doc (archetype, learning loop, mentor frame, OS check) and config file. Also optionally creates a custom panel-review skill using research agents on named advisors. Run once before the first /ramp session. Re-run any time to update config. triggers: - setup ramp - configure ramp - setup-ramp allowed-tools: - Bash - Read - Write - Edit - Agent - AskUserQuestion - WebSearch --- # /setup-ramp ## Step 0 — Check for existing config Run: ```bash [ -f ~/.claude/ramp-config.yaml ] && echo "EXISTS" || echo "NEW" [ -f ~/.claude/ramp-context.md ] && echo "CONTEXT_EXISTS" || echo "CONTEXT_NEW" ``` If config exists, ask via AskUserQuestion: ``` Question: A config already exists. What do you want to do? Options: A) Update it — go through the questions, keep existing answers as defaults B) Start fresh — overwrite everything C) Just update context — redo the context interview, keep config as-is ``` Load existing config and context if updating (so you can show current values as defaults). --- ## Phase 1 — Context interview **Goal:** Compose `~/.claude/ramp-context.md` — the equivalent of a personal founder-flywheel doc. This is what the skill reads at the start of every session to understand the user's mental model. Ask each question as plain text, one at a time. Wait for the full response before proceeding. Show existing values as "(current: ...)" when updating. **Q1 — Operating pillars:** > "What are your 3-5 core operating pillars as a founder — the things you want to track and improve every week? These are the axes your whole practice runs on. > > Examples: deployment reps / market exposure, mentor relationships, public brand for fundraising, health and operating rhythm, compound tooling and engineering. > > Name yours and say one sentence about why each one matters to you right now." **Q2 — Mentor relationship frame:** > "How do you think about learning from advisors and mentors? Three parts: > > 1. Mindset — what's your posture going into those conversations? What are you trying to get from them? Say how you actually think about it, doesn't need to be polished. > > 2. Intended cadence — how often do you want to be in contact with your primary mentor? (e.g. weekly, every 2 weeks, monthly, quarterly, or 'ad-hoc' if you prefer to reach out only when something specific comes up) > > 3. Modality — how do you prefer to communicate with them? (e.g. async written updates, voice memos, scheduled calls, in-person, Slack DMs, email threads) And what style — terse status, narrative paragraph, bullet list, voice ramble?" Parse the response into three named fields. Handle edge cases: - **If cadence is given as a phrase, not a duration** (e.g. "when it feels right", "regularly", "often"): push once: > "Give me a rough interval in weeks, or say 'ad-hoc' if there's no intended rhythm." Accept the second answer as-is. - **If cadence is 'ad-hoc' or equivalent**: record cadence as `ad-hoc` and set `default_send_window` to "within 5 business days of deciding to send." - **For numeric cadences**: derive `default_send_window` as roughly one-third of the cadence interval, capped at 5 business days: - Weekly cadence → "by end of this week" (default Friday) - Biweekly → "within 5 business days" - Monthly → "within 5 business days" - Quarterly → "within 2 weeks" - **If modality is vague** (e.g. "messages", "we talk"): push once: > "Async or synchronous? Written or voice?" Then accept. - **If style is not given**: infer from modality (voice memo → "voice ramble", Slack DM → "terse bullets", email → "narrative paragraph") and confirm: > "Defaulting style to [X] — override?" Accept either confirmation or correction. **Q3 — Founder trajectory:** > "Where are you right now? Three modes — pick the one that fits: > > **A) Compound engineer** — not founding any time soon. Building skills, technical leverage, and optionality for a future founding event. > > **B) Thesis mode** — clear on a domain or problem space, no specific idea crystallized yet. Sharpening the thesis. > > **C) Idea mode** — something concrete to build toward, with at least a rough timeline. > > Which fits? And say one sentence about what's actually constraining you right now — time, money, conviction, co-founders, or something else." Based on the archetype answer, ask one conditional follow-up (only this one, then move on): - **Compound**: no follow-up on timeline. Record only the structural constraint. - **Thesis**: > "What's the domain, and what's the single biggest thing you don't yet know about it?" - **Idea**: > "What's your rough timeline, and what's the first concrete milestone — not 'launch a startup' but the next specific gate?" Then, regardless of archetype, ask: > "Which of these do you actively want to skip tracking in your weekly sessions? Technical depth, compound engineering, and mentor relationships are always on. > > - **Sales / customer discovery** — direct conversations with potential customers, field reps, ground-truth market signal > - **Marketing / brand** — public publishing, audience building, brand for hiring or fundraising > - **Fundraising / investor relationships** — investor network, fundraising readiness, pitch cadence > > Name any to drop, or 'none' to track all." **Q4 — Work context:** > "Describe your work context in 2-3 sentences: your job or main project right now, what you're building or learning, and what the ground-truth learning surface looks like — where do you actually get signal about customers, markets, or the deployment gap?" This feeds into the learning loop synthesis used to frame Q2 in every ramp session. --- After all 4 answers, synthesize into `~/.claude/ramp-context.md`: ```markdown --- last_updated: YYYY-MM-DD --- # Ramp — Personal Context ## Operating Pillars [Synthesize from Q1 — list each pillar with a 1-2 sentence description of why it matters. Use their words where possible.] ## Mentor Relationship Frame [Synthesize from Q2 mindset answer — describe their posture for advisor relationships. First-person-adjacent prose, not a list.] **Intended cadence:** [e.g. "every 2 weeks" | "ad-hoc"] **Default send window:** [e.g. "within 5 business days" | "by end of week"] **Preferred modality:** [e.g. "async written update via email"] **Style:** [e.g. "bullet list, terse"] ## Founder Trajectory **Archetype:** [compound | thesis | idea] **Structural constraint:** [one sentence — what's actually constraining them right now] [Include only if thesis: **Thesis domain:** [domain] | **Biggest unknown:** [unknown]] [Include only if idea: **Timeline:** [rough timeline] | **First milestone:** [specific milestone]] [One sentence synthesizing current situation and direction. Factual, specific.] ## Work Context [Q4 answer, lightly edited for clarity. 2-4 sentences. This is read verbatim to frame Q2 in each session.] ## Weekly Focus **Skipped axes:** [e.g. "Marketing / brand, Fundraising / investor relationships" | "none"] **Learning loop:** [synthesize from archetype + work context — the phrase interpolated into Q2 each session. Examples: "compound engineering and tooling" | "product iteration with early users" | "deployment loop from real customer conversations" | "content and distribution experiments" Only ask the user explicitly if synthesis is ambiguous.] ``` --- ## Phase 2 — Config interview **Goal:** Write `~/.claude/ramp-config.yaml`. --- ### OS check trial Before any file path questions, run a quick trial of the default OS check so the user can shape it from real answers, not abstractions. > "Before we get into file paths, let me show you what your weekly OS check looks like. Answer these quickly." Ask the three default items as plain questions, one at a time: 1. > "Health pillars: how did sleep, fitness, mindfulness/journaling, and a no-electronics morning hour go this week? What held and what slipped?" 2. > "Compound engineering: what did you build or improve this week that increases the rate of your future work?" 3. > "Growth metric: what single number are you tracking right now that tells you founder-readiness is moving?" After all three answers, present the default item list and offer to prune or add: > "Here's what your OS check will look like each week: > - Health pillars (sleep, fitness, mindfulness/journaling, no-electronics morning) > - Compound engineering > - Growth metric > > Anything to drop? Anything to add that came up in your answers?" Accept their edits and store the final list as `os_check_items` in config. Each item is a short label (e.g. "Health pillars", "Compound engineering", "Reading pace"). If they keep everything and add nothing, store the three defaults. --- ### File paths — single vault question Ask only ONE path question: > "Andrej Karpathy calls this 'manipulating knowledge' — using an LLM not just to chat, but to build and maintain a personal knowledge base over time. > > A vault is that knowledge base: a local folder of plain-text markdown files — notes, wiki pages, ideas — that compound as you work. > > Do you have one already? Paste the path, or press enter and I'll create one at `~/Documents/Notes`." Save the answer (or `~/Documents/Notes`) as `vault_dir`. That's the only path question. All other locations are derived automatically. --- **Q5 — Primary mentor:** > "Who's your primary mentor or advisor — the person whose judgment on your founder progress matters most right now? (name, or 'none')" If they name someone, save as `mentor_name`. If 'none', leave blank. --- **Q6 — Distribution / brand mode:** > "How do you want Q4 of each weekly session to work — the axis that tracks your distribution-building? > > **A) Both** — public publishing (posts, threads, blog) AND private network moves (outreach, intros, relationships) both count as valid answers. > **B) Brand only** — public publishing only. Did the artifact ship? > **C) Network only** — networking flywheel only. Did the relationship move happen? > **D) Off** — skip this axis entirely each session. > > Pick one, or press enter for the default (Both)." Save as `q4_mode` (values: `both` | `brand` | `network` | `off`). Default: `both`. --- ### Autodetect paths from vault structure After getting `vault_dir` and `mentor_name`, scan the vault to resolve all file paths automatically. Run: ```bash # Check if vault exists [ -d "[vault_dir]" ] && echo "VAULT_EXISTS" || echo "VAULT_NEW" # List all markdown files for semantic inspection find "[vault_dir]" -maxdepth 5 -name "*.md" 2>/dev/null # List all directories find "[vault_dir]" -maxdepth 4 -type d 2>/dev/null ``` Read the file listing and use your judgment to identify which file is most likely serving each role. You are looking for **~85% semantic confidence** — don't require exact names, but require that the file's name, path, and/or surrounding directory plausibly matches the function. Skim any ambiguous files to check their structure before deciding. Detection heuristics by role (examples, not exhaustive): | Config field | What you're looking for | |---|---| | `mentor_notes_file` | A file in or near a `people/` directory whose name resembles the mentor's name — slug, first name, last name, initials, or common nickname. Should look like a profile/notes page, not a log. | | `mentor_updates_file` | A file near the mentor notes file whose name suggests a log, updates, cadence, or sent messages — e.g. `mike-updates.md`, `leblanc-log.md`, `mentor-log.md`. | | `open_questions_file` | Any file whose name or path suggests unanswered questions, open loops, or unknowns — e.g. `open-questions.md`, `unknowns.md`, `questions.md`, `things-to-figure-out.md`. | | `current_actions_file` | Any file whose name suggests current priorities, this week's focus, active todos, or a top-N list — e.g. `current-actions.md`, `this-week.md`, `top3.md`, `priorities.md`. | | `crm_file` | Any file that looks like a contact list, relationship tracker, or people CRM — e.g. `crm.md`, `contacts.md`, `people.md`, `network.md`. | | `session_output_dir` | A directory that looks like it holds dated journal entries or session notes — e.g. `raw/journal`, `journal/`, `sessions/`, `weekly/`. | | `ideas_dir` | A directory that looks like it holds ideas, sparks, or captured concepts — e.g. `raw/ideas`, `ideas/`, `sparks/`, `inbox/`. | If confidence is below ~85% for a given field, do not guess — fall back to the default path and mark it "will create." Default paths if not detected: | Config field | Default | |---|---| | `mentor_notes_file` | `[vault_dir]/wiki/people/[slug].md` | | `mentor_updates_file` | `[vault_dir]/wiki/people/[slug]-updates.md` | | `open_questions_file` | `[vault_dir]/wiki/open-questions.md` | | `current_actions_file` | `[vault_dir]/wiki/current-actions.md` | | `crm_file` | `[vault_dir]/wiki/people/crm.md` | | `session_output_dir` | `[vault_dir]/raw/journal` | | `ideas_dir` | `[vault_dir]/raw/ideas` | Ingestion targets always default to: mentor_contact → `mentor_updates_file`, new_todos → `current_actions_file` (if those files are configured). --- ### Single confirmation before writing Before creating any files or writing config, display a summary table: > "Here's what I'll set up — derived from your vault. Say 'go' to proceed, or correct any path." ``` vault_dir: [vault_dir] mentor_notes: [path] [DETECTED | will create] mentor_updates: [path] [DETECTED | will create | —] open_questions: [path] [DETECTED | will create] current_actions: [path] [DETECTED | will create] crm: [path] [DETECTED | will create] session_output_dir: [path] [DETECTED | will create] ideas_dir: [path] [DETECTED | will create] ``` Wait for "go" (or "yes", "looks good", "proceed"). If they correct a path, update it and re-display the table. Only after confirmation: create missing files/directories and write config. **File templates for "will create" cases:** `mentor_notes_file`: ```markdown --- type: person last_updated: YYYY-MM-DD --- # [Mentor Name] ## Background ## How They Think / Mental Models ## Key Advice and Framings ## Learnings Log ``` `mentor_updates_file`: ```markdown --- type: reference last_updated: YYYY-MM-DD --- # [Mentor Name] — Update Log ## Entries *(most recent first)* ``` `open_questions_file`: ```markdown # Open Questions ## Fundraising & Investors ## Product & Market ## Operating & Personal ## Mentor & Network ``` `current_actions_file`: ```markdown # Current Actions ## Top 3 This Week *(updated after each /ramp session)* ## Open Todos ``` `crm_file`: create as an empty markdown file with a `# Contacts` heading. --- After all config questions, write `~/.claude/ramp-config.yaml`: ```yaml # Ramp — Personal Config # Generated by /setup-ramp on YYYY-MM-DD # Edit any field directly. Blank values disable that feature. vault_dir: "[vault_dir]" mentor_name: "[mentor_name or blank]" mentor_notes_file: "[path or blank]" mentor_updates_file: "[path or blank]" open_questions_file: "[path or blank]" current_actions_file: "[path or blank]" crm_file: "[path or blank]" session_output_dir: "[path]" ideas_dir: "[path]" q4_mode: "[both | brand | network | off]" # default: both os_check_items: - "[item 1]" - "[item 2]" - "[item 3]" panel_review_skill: "" # filled in Phase 3 if created ingestion_targets: [- file: "..." / on: "..." for each configured target] ``` --- ## Phase 3 — Founder-critique setup (optional) Ask as plain text: > "Do you want to set up a founder-critique skill — a way to get your ramp sessions reviewed from multiple advisor perspectives? I'll research your chosen advisors and write a custom skill for it. (yes / no)" If yes: **Step 1 — Name your panel:** > "Name 3 people — real founders, investors, or thinkers — whose judgment you most want in the room. For each one, say one sentence about the lens they bring. > > Example: 'Paul Graham — cuts through narrative to what's actually happening. Naval Ravikant — thinks in leverage and long games. Jensen Huang — demands the hardest thing be done first.'" **Step 3 — Research agents:** Spawn one background research agent per named person (in parallel). Each agent: 1. Searches for the person's known mental models, public writing, areas of expertise, and characteristic critiques 2. Finds 3-5 specific, quotable framings or patterns from their public record 3. Returns a brief profile: background (2 sentences), their lens as a founder/advisor, their characteristic critique pattern, and 3-5 specific framings or quotes Collect all three agent results before proceeding. **Step 4 — Write the panel-review skill:** Create `~/.claude/skills/founder-critique/SKILL.md` modeled on the /panel-review structure: ```markdown --- name: founder-critique preamble-tier: 0 version: 1.0.0 description: | [N]-advisor panel critique of ramp progress. Loads profiles for [names], then applies each advisor's critique lens to recent ramp sessions. triggers: - panel review - founder-critique - advisor panel allowed-tools: - Bash - Read - Write --- # /founder-critique ## Step 1 — Load sessions Load the N most recent ramp session files from [session_output_dir]: ```bash ls -t [session_output_dir]/*ramp-session.md 2>/dev/null | head -3 ``` Read each. Also read ~/.claude/ramp-context.md for current context. ## Step 2 — Individual critiques For each advisor, output a `## [Name]` section. **Format:** **Part 1 — Thought process (first-person, present tense):** The advisor reads through the sessions as they go, narrating reactions. Reference specific moments. Show reasoning before verdict. 3-5 paragraphs. **Part 2 — Verdict:** - **Score (1-5)** with one sentence anchored to specific evidence - **The one question they'd leave with** — their single sharpest forcing question --- ### [Advisor 1 Name] **Background:** [2-sentence background from research] **Lens:** [how they think about founder progress — from research] **Characteristic critique pattern:** [what they typically push on — from research] **Key framings to apply:** [3-5 specific framings or quotes from research, used to shape the critique voice] **Voice calibration:** [2-3 sentences on how they sound — their register, sentence length, what they're merciless about] --- ### [Advisor 2 Name] [Same structure] --- ### [Advisor 3 Name] [Same structure] --- ## Step 3 — Panel synthesis Always output `## Panel Synthesis` with: 1. **Where they agree** — highest-confidence signal 2. **The sharpest disagreement** — where their frameworks diverge 3. **The one action they'd all endorse** — most convergent next step ## Step 4 — Save output Create `[session_output_dir]/../YYYY-MM-DD-founder-critique.md` with the full critique output. Commit: ```bash git add [output file] git commit -m "chore: founder-critique session $(date +%Y-%m-%d)" ``` ``` Update `panel_review_skill` in `~/.claude/ramp-config.yaml` to `founder-critique`. --- ## Phase 4 — Finish Print: > "Setup complete." > "" > "Run `/ramp` to start your first session." > "" > "Files written:" > " - Context doc: ~/.claude/ramp-context.md" > " - Config: ~/.claude/ramp-config.yaml" > [if panel-review created] " - Panel review skill: ~/.claude/skills/founder-critique/SKILL.md" > "" > "Edit either file any time — changes take effect immediately on next session." > "" > [if any files were created by "create one" choices] "Files created in your vault:" > [list each created file]