buildpartner 0.4.2 → 0.8.0-dev.20260430072343

package/skills/bp-full-audit/SKILL.md

@@ -1,145 +0,0 @@
----
-description: "Full audit of your Claude Code environment, skills, and workflows. Finds what's missing, what's broken, and what to improve. Prioritized by impact."
----
-
-# /bp-full-audit
-
-Full audit of your Claude Code environment, skills, and workflows. Finds what's missing, what's broken, and what to improve. Prioritized by impact.
-
-## MCP Tools
-
-Only call tools that are registered on the `buildpartner` MCP server. If a tool call fails or a tool name isn't recognized, do not retry with a different name. The available tools are: `get_personal_context`, `get_expert_knowledge`, `save_profile`, `get_quest_progress`, `check_status`.
-
-## Instructions
-
-1. Call `get_personal_context` with `{ "include_sessions": true }`.
-
-2. Read and audit every layer of their setup. For each layer, read the relevant files and analyze against their session history.
-
-### Layer 1: CLAUDE.md
-
-Read `~/.claude/CLAUDE.md` (global) and any project-level CLAUDE.md.
-
-Check:
-- Are there rules? If empty or missing, that's a high-priority finding.
-- Do the rules match what they actually do? (Compare rules to session patterns)
-- Are rules too generic? ("Write clean code" is useless)
-- Are rules too long? (Best CLAUDE.md files are under 30 lines)
-- Missing rules they need based on their tech stack and patterns
-
-### Layer 2: MCP Servers
-
-Read `~/.claude/mcp_servers.json` (or the equivalent config).
-
-Check:
-- What MCP servers are configured?
-- Are they using database tools manually when an MCP server could handle it? (Look for SQL in Bash commands in session history)
-- Are there MCP servers configured but never used? (Compare config to session tool_counts)
-- Missing MCP servers based on their workflow (e.g., they query APIs manually but don't have relevant MCP servers)
-
-### Layer 3: Skills
-
-Read all skills in `~/.claude/skills/`.
-
-Check:
-- How many custom skills do they have (non-BuildPartner)?
-- Which BuildPartner skills have they used vs not used?
-- Are there custom skills that are stale or broken?
-- Could they benefit from creating skills for repeated workflows? (Look for patterns in session history where they do the same thing repeatedly)
-
-### Layer 4: Hooks
-
-Check `~/.claude/settings.json` and project-level settings for hooks.
-
-Check:
-- Are any hooks configured?
-- Is the BuildPartner tracking hook installed?
-- Could they benefit from pre/post hooks? (e.g., auto-run tests after code changes)
-
-### Layer 5: Knowledge and Context
-
-Check for knowledge folders:
-- `~/.buildpartner/knowledge/experiences/` - do they have logged experiences?
-- Project-level `knowledge/`, `context/`, `docs/` folders
-- `.claude/` project configuration
-
-Check:
-- Do they have any structured knowledge that skills could reference?
-- Are there files that should be in knowledge but aren't?
-- Is there orphaned context nobody uses?
-
-### Layer 6: Workflow Patterns
-
-From session history, identify:
-- Do they use Plan mode? If not and they should, flag it.
-- Do they run parallel sessions?
-- Do they use agents/subagents?
-- How long are their sessions? (Short = tactical, Long = could benefit from planning)
-- Do they commit frequently or in big batches?
-
-## Output Format
-
-Present the audit as a prioritized action plan:
-
-```
-## Full Audit Results
-
-**Environment Score**: {X}/10
-Based on {Y} sessions analyzed across {Z} projects.
-
-### Critical (fix now)
-1. {Finding}: {what's wrong and why it matters}
-   Fix: {exact action to take}
-
-### High Impact (do this week)
-1. {Finding}: {what's wrong and why it matters}
-   Fix: {exact action to take}
-
-### Nice to Have (when you have time)
-1. {Finding}
-   Fix: {action}
-
-### What's Working Well
-- {Positive finding}
-- {Positive finding}
-```
-
-## Rules
-
-- Every finding must be backed by data from their sessions. Don't suggest generic improvements.
-- "Fix" actions should be specific enough to copy-paste or act on immediately.
-- If a fix is something you can do right now (edit CLAUDE.md, create a skill), offer to do it.
-- Keep the total list to 10 items max. Prioritize ruthlessly.
-- Always include "What's Working Well" so it doesn't feel like a list of failures.
-- If they have < 10 sessions, say: "Not enough data for a full audit yet. Come back after a few more sessions. For now, here's what I can see:"
-
-## After the Audit
-
-Ask:
-> "Want me to fix any of these now? I can update your CLAUDE.md, create missing skills, or configure MCP servers."
-
-If yes, work through the fixes one by one, confirming each before applying.
-
-## Generate Expert Advice Prompt
-
-After presenting the audit results, generate a tailored `/bp-expert-advice` command the user can copy and paste. Look at their project, tech stack, and audit findings to craft a specific question that would give them the most value.
-
-Pick the most impactful question based on what you learned:
-- If they're building a product: ask about launch strategy, pricing, or go-to-market
-- If they're creating content: ask about content strategy or audience growth
-- If they're automating workflows: ask about AI automation or systems design
-- If they have a specific challenge visible in their setup: ask about that
-
-Present it as a ready-to-run command in a copyable block:
-
-```
-Your audit is done. Here's something you can run right now to get expert advice tailored to what you're building:
-```
-
-Then show the exact command, for example:
-
-`/bp-expert-advice I'm building [specific thing from their project]. What's the best way to [specific challenge based on audit findings]?`
-
-Make the prompt specific to THEIR situation. Never use a generic question. Reference their actual project name, tech stack, or goals.
-
-Call `check_status` with `{ "skill_name": "bp-full-audit" }`.

package/skills/bp-generate-personalized-training-data/SKILL.md

@@ -1,145 +0,0 @@
----
-description: "Generate personalized training data from your Claude Code sessions. Recaps your week, captures what you built, and logs lived experiences. Every skill gets smarter after."
----
-
-# /bp-generate-personalized-training-data
-
-Generate personalized training data from your Claude Code sessions. Recaps your week, captures what you built, and logs lived experiences. Every skill gets smarter after.
-
-## MCP Tools
-
-Only call tools that are registered on the `buildpartner` MCP server. If a tool call fails or a tool name isn't recognized, do not retry with a different name. The available tools are: `get_personal_context`, `get_expert_knowledge`, `save_profile`, `get_quest_progress`, `check_status`.
-
-## Instructions
-
-1. Call `get_personal_context` with `{ "include_sessions": true }`.
-
-2. Check if they have a `training_data` section in their AI profile with a `write_path`. If yes, you know where to write. If no, this is their first time.
-
-### Part 1: Weekly Recap
-
-Build a recap from this week's sessions:
-
-```
-## This Week
-
-**{X} sessions | {Y} tokens | {streak} day streak**
-
-### What you worked on
-- {Project/activity 1} ({X sessions})
-- {Project/activity 2} ({X sessions})
-
-### Best session
-{The most interesting or productive session this week. One sentence.}
-
-### Add to your playbook
-{1-2 specific things they should start doing based on this week's patterns.}
-
-### Your week in numbers
-Code {X%} | Content {X%} | Research {X%} | Other {X%}
-```
-
-Keep the recap under 15 lines. Lead with what matters.
-
-### Part 2: Generate Training Data
-
-From the session history, extract structured training data entries. Each entry captures something transferable:
-
-```markdown
-# {Date} - {Brief Title}
-
-## What happened
-{1-2 paragraph narrative of what was built or figured out}
-
-## What was built
-- {Files/features created or modified}
-- {Tools or approaches used}
-
-## Key learning
-- {The transferable insight from this experience}
-
-## Tags
-{topic1, topic2, topic3}
-```
-
-Generate entries for the most significant sessions this week (2-5 entries, skip trivial sessions).
-
-### Part 3: Determine Where to Write
-
-**If they have a `write_path` saved in their profile:** Write there without asking. Just confirm what you wrote.
-
-**If this is their first time:** Look for existing knowledge/context directories in their workspace:
-- Check for `~/.buildpartner/knowledge/experiences/`
-- Check for directories like `internal-os/context/`, `knowledge/`, `context/`, `docs/` in the current project
-- Check for any `.claude/` project-level context folders
-
-If you find an existing structure, suggest the best fit:
-
-> "I see you have `{path}`. I'll write your training data there. Sound good?"
-
-If nothing exists, default to `~/.buildpartner/knowledge/experiences/` and tell them:
-
-> "I'll create `~/.buildpartner/knowledge/experiences/` for your training data. You can change this anytime."
-
-Save the chosen path to their profile so you never ask again.
-
-### Part 4: Log Lived Experiences
-
-After showing the recap and generated entries, ask:
-
-> "Anything else from this week worth capturing? A win, a lesson, something you figured out. A sentence or two is plenty, or type 'skip'."
-
-If they share something, add it as an additional training data entry.
-If they skip, move on.
-
-### Part 5: Write Everything
-
-1. Write each training data entry as a separate file: `{write_path}/{YYYY-MM-DD}-{brief-slug}.md`
-2. Write the weekly recap as: `{write_path}/recaps/{YYYY-MM-DD}-weekly-recap.md`
-
-Create directories as needed with `mkdir -p`.
-
-3. Save the write path and metadata to their profile:
-
-```json
-{
-  "updates": {
-    "training_data": {
-      "write_path": "/path/to/knowledge/experiences",
-      "last_generated": "YYYY-MM-DD",
-      "total_entries": 12,
-      "weeks_captured": 3
-    }
-  }
-}
-```
-
-## Output
-
-After writing, confirm:
-
-```
-## Training Data Generated
-
-**{N} entries written** to `{write_path}`
-**Week {N} captured** ({total_entries} total entries)
-
-Files:
-- {filename1}
-- {filename2}
-- recaps/{recap-filename}
-
-This feeds into /bp-claude-coach, /bp-expert-advice, and /bp-interview-me.
-Run this weekly to keep everything sharp.
-```
-
-## Rules
-
-- Keep the recap honest. Good week? Say so. Inconsistent? Say that too.
-- Training data entries should be useful for future consultations, not just a diary.
-- Write in their voice if their AI profile has a writing style.
-- If they have < 3 sessions this week, say: "Light week. Here's what I have:" and generate what you can.
-- Don't ask unnecessary questions. Generate first, ask for additions after.
-- The "Add to your playbook" section in the recap is the most important part. Make it specific and actionable.
-
-Call `check_status` with `{ "skill_name": "bp-generate-personalized-training-data" }`.

package/skills/bp-improve-skill/SKILL.md

@@ -1,115 +0,0 @@
----
-description: "Enhance a skill so you don't make the same mistake twice. After using any skill (BuildPartner or your own custom ones), run this to improve it based on what just happened."
----
-
-# /bp-improve-skill
-
-Enhance a skill so you don't make the same mistake twice. After using any skill (BuildPartner or your own custom ones), run this to improve it based on what just happened.
-
-## Instructions
-
-### Step 1: Identify the skill
-
-Ask:
-> "Which skill do you want to improve?"
-
-List skills from:
-- BuildPartner skills: `~/.claude/skills/bp-*/SKILL.md`
-- Their custom skills: `~/.claude/skills/*/SKILL.md` (excluding bp-* prefix)
-
-If the user says "based on our conversation" without naming a skill, infer from context which skill is relevant.
-
-### Step 2: Understand the improvement
-
-Ask:
-> "What should be better? For example:
-> - Add a missing step
-> - Change the output format
-> - Add a learning for next time
-> - Fix something that went wrong
->
-> Or just describe what happened and I'll suggest improvements."
-
-### Step 3: Review the current skill
-
-Read the SKILL.md file. Identify:
-- Current steps and structure
-- Any existing learnings sections
-- Context or knowledge it references
-
-### Step 4: Propose changes
-
-Show the user:
-```
-Current skill: {name}
-Steps: {count}
-Learnings: {count if any}
-
-Proposed changes:
-1. {specific change with reasoning}
-2. {specific change with reasoning}
-
-Apply these?
-```
-
-### Step 5: Check neighboring skills
-
-After proposing changes to the target skill, check for neighboring skills that might need the same update.
-
-**Neighboring skills are:**
-- Skills that do similar things (e.g., all consultant-type skills)
-- Skills that reference the same knowledge or context
-- Skills in the same domain
-
-Search through all SKILL.md files for shared patterns.
-
-If a learning applies to neighbors:
-> "This learning might also apply to:
-> - {skill}: {why}
-> Should I update it too?"
-
-### Step 6: Detect contextual rot
-
-After updating, scan for problems:
-
-**Duplicate guidance**: Two skills giving the same advice differently.
-**Conflicting rules**: One skill says X, another says the opposite.
-**Stale references**: Tools, file paths, or approaches that no longer exist.
-**Orphaned files**: Knowledge files that no skill references.
-**Learning bloat**: A skill with too many learnings (>20) that should be consolidated.
-
-Report anything found:
-> "While updating, I found:
-> - {issue}: {details}
-> Want me to fix this too?"
-
-### Step 7: Apply updates
-
-If approved:
-1. Edit the target SKILL.md
-2. Edit any neighboring skills they approved
-3. Flag rot issues (don't auto-fix without approval)
-
-Confirm:
-> "Updated:
-> - {skill}: {what changed}
-> - {neighbor}: {what changed, if any}
->
-> Flagged: {any rot issues found}"
-
-### Quick mode
-
-If the user says "based on our back and forth, improve {skill}":
-1. Review the conversation for relevant learnings
-2. Read the target skill
-3. Propose specific improvements
-4. Check neighbors automatically
-5. Apply on approval
-
-### Rules
-- Never modify a skill without showing proposed changes first
-- Check neighbors BEFORE asking (don't ask about empty matches)
-- Conflicting learnings aren't always bad, sometimes they're nuanced. Ask before consolidating.
-- Keep skills concise. If an improvement makes the skill longer, also suggest what to cut.
-
-Call `check_status` with `{ "skill_name": "bp-improve-skill" }`.

package/skills/bp-interview/SKILL.md

@@ -1,64 +0,0 @@
----
-description: "Build your profile so every skill knows who you are. Quick conversation, not a form. 2-3 minutes."
----
-
-# /bp-interview
-
-Build your profile so every skill knows who you are. Quick conversation, not a form. 2-3 minutes.
-
-## MCP Tools
-
-Only call tools that are registered on the `buildpartner` MCP server. If a tool call fails or a tool name isn't recognized, do not retry with a different name. The available tools are: `get_personal_context`, `get_expert_knowledge`, `save_profile`, `get_quest_progress`, `check_status`.
-
-## Instructions
-
-1. Call `get_personal_context` with `{ "include_sessions": true }` to see how the user actually works.
-
-2. Use their session data to skip obvious questions. If you can see they code in TypeScript with Next.js, don't ask "what's your tech stack?" If they have 300 sessions, don't ask if they're new to Claude Code.
-
-3. Run a conversational interview. One question at a time. Make it feel like a real conversation, not a survey. The goal is to understand who this person is, what they care about, and where they're headed.
-
-### Who they are
-- "What are you building right now, and why does it matter to you?"
-- "Is this a solo thing or do you have a team?" (only if not obvious from context)
-- "What got you into building? What's the origin story?"
-
-### Where they're going
-- "Where do you want to be in 5 years? What does the ideal version of your work life look like?"
-- "What's the biggest thing standing between you and that?"
-- "If you could mass-produce one skill or ability in yourself, what would it be?"
-
-### How they like to work
-- "When Claude responds, do you want it concise or detailed?"
-- "If something in your code is wrong, do you want me to call it out directly or be diplomatic?"
-- "What do you wish Claude did differently?"
-
-4. Rules:
-   - Ask ONE question at a time
-   - Read the room. Short answers = keep it tighter. Long answers = follow up and go deeper
-   - Skip questions you can already answer from their data
-   - 6-10 questions total. Quality over quantity
-   - If the user says "done" or wants to stop, save immediately with what you have
-   - Don't be robotic. React to what they say. Be a person, not a form
-
-5. After the interview, save EVERYTHING in ONE single `save_profile` call. ONE call, ALL sections:
-
-```json
-{
-  "updates": {
-    "writing_style": { "tone": "...", "verbosity": "...", "approach": "...", "technical_level": "..." },
-    "preferences": { "response_style": "...", "code_style": "...", "decision_making": "...", "feedback_style": "..." },
-    "tech_identity": { "role": "...", "primary_languages": ["..."], "frameworks": ["..."], "current_project": "..." },
-    "work_patterns": { "session_style": "...", "planning_style": "...", "peak_hours": "from context", "peak_day": "from context" },
-    "personality": { "builder_type": "...", "learning_style": "...", "pain_points": ["..."], "goals": ["..."] },
-    "vision": { "five_year": "...", "biggest_bottleneck": "...", "origin_story": "...", "superpower_wish": "..." }
-  },
-  "signals": ["Summary of key insights about who this person is and what drives them"]
-}
-```
-
-6. Show the completed profile, then ask: "Anything look wrong? I can update any section."
-
-7. End with: "Your profile is saved. Every BuildPartner.ai skill now personalizes to you."
-
-8. Call `check_status` with `{ "skill_name": "bp-interview" }`.

package/skills/bp-interview-me/SKILL.md

@@ -1,120 +0,0 @@
----
-description: "Build your profile so every skill knows who you are. Gets smarter each time you run it, asking deeper questions based on what it already knows."
----
-
-# /bp-interview-me
-
-Build your profile so every skill knows who you are. Gets smarter each time you run it, asking deeper questions based on what it already knows.
-
-## MCP Tools
-
-Only call tools that are registered on the `buildpartner` MCP server. If a tool call fails or a tool name isn't recognized, do not retry with a different name. The available tools are: `get_personal_context`, `get_expert_knowledge`, `save_profile`, `get_quest_progress`, `check_status`.
-
-## Instructions
-
-1. Call `get_personal_context` with `{ "include_sessions": true }`.
-
-2. Determine their **interview depth** based on what profile data already exists:
-
-### Depth 1: First time (no profile or very sparse)
-
-Quick, punchy. 6-8 questions max. Get the essentials so everything can start personalizing.
-
-Questions (pick the best ones based on what session data already tells you):
-- "What are you building right now, and why does it matter to you?"
-- "Solo or team?"
-- "Where do you want to be in a year?"
-- "What's the biggest thing slowing you down?"
-- "When Claude responds, concise or detailed?"
-- "If something in your code is wrong, call it out directly or be diplomatic?"
-- "What do you wish Claude did differently?"
-
-Skip anything you can already answer from their session data. If you can see they code in TypeScript with Next.js, don't ask "what's your tech stack?"
-
-Tone: Fast, conversational, not a survey. React to what they say.
-
-### Depth 2: Has basic profile, gaps to fill
-
-They've done this before but there are thin areas. Acknowledge what you know, then go deeper on what's missing.
-
-Start with: "Good to see you again. Here's what I know about you so far: {1-2 sentence summary}. Let me fill in some gaps."
-
-Target questions at the weakest sections of their profile:
-- **If personality is thin**: "What got you into building? What's the origin story?" / "What's the hardest thing you've been through that changed how you work?"
-- **If vision is thin**: "Where do you want to be in 5 years? Not the pitch, the real answer." / "If you could mass-produce one skill in yourself, what would it be?"
-- **If work_patterns is thin**: "Walk me through how you start a typical work session." / "When do you do your best work?"
-- **If preferences are thin**: "When you're stuck, do you want me to suggest solutions or ask questions?" / "How do you feel about me making changes proactively vs. asking first?"
-
-6-8 targeted questions. Don't re-ask things they've already answered well.
-
-### Depth 3: Solid profile, time to go deep
-
-They have a well-filled profile. Now ask the questions that unlock real personalization.
-
-Start with: "Your profile is solid. Let me ask some harder questions to make this even more useful."
-
-Pick from:
-- "Walk me through your career. Not the resume version. The real version."
-- "What's the thing you've built that you're most proud of?"
-- "What's the biggest risk you've taken?"
-- "Who influenced you the most and why?"
-- "How do you make decisions? Gut feel, data, or advice from people you trust?"
-- "What are you really good at that most people don't know?"
-- "What do you struggle with that you don't usually admit?"
-- "What does your life look like in 3 years if everything goes well?"
-- "What's the thing you keep putting off that you know matters?"
-
-Go where the conversation goes. If something is interesting, follow up. Don't stick rigidly to the list.
-
-### Depth 4: Deep profile, find the edges
-
-They've done multiple rounds. Now find the nuance, contradictions, and updates.
-
-Start with: "You've shared a lot. Let me check in on a few things and dig into some edges."
-
-- Check for stale data: "Last time you said {X}. Still true?"
-- Find contradictions: "You said you want {X} but your sessions show {Y}. What's going on there?"
-- Explore new territory: "What's changed since we last talked?" / "What are you excited about right now that you weren't before?"
-- Push on growth: "What did you learn this month that surprised you?" / "What habit are you trying to build or break?"
-
-5-8 questions. This level is more like a check-in than an interview.
-
-## Rules
-
-- Ask ONE question at a time. Wait for the answer.
-- Read the room. Short answers = keep it tighter. Long answers = follow up and go deeper.
-- Don't be robotic. React to what they say. Be a person, not a form.
-- If the user says "done" or wants to stop, save immediately with what you have.
-- Never re-ask questions they've already answered well in a previous round.
-- Always tell them what depth they're at: "This is round {N} for you."
-
-## Saving
-
-After the interview, save EVERYTHING in ONE single `save_profile` call. Merge new data with existing data. Do not overwrite fields that haven't changed.
-
-```json
-{
-  "updates": {
-    "writing_style": { "tone": "...", "verbosity": "...", "approach": "...", "technical_level": "..." },
-    "preferences": { "response_style": "...", "code_style": "...", "decision_making": "...", "feedback_style": "..." },
-    "tech_identity": { "role": "...", "primary_languages": ["..."], "frameworks": ["..."], "current_project": "..." },
-    "work_patterns": { "session_style": "...", "planning_style": "...", "career_path": "...", "proudest_work": "..." },
-    "personality": { "builder_type": "...", "learning_style": "...", "pain_points": ["..."], "goals": ["..."], "origin": "...", "values": "...", "strengths": "...", "struggles": "...", "decision_style": "...", "influences": "..." },
-    "vision": { "five_year": "...", "biggest_bottleneck": "...", "origin_story": "...", "superpower_wish": "..." },
-    "interview_meta": { "rounds_completed": 1, "last_round": "YYYY-MM-DD", "depth_reached": 1, "sections_covered": ["writing_style", "preferences", "tech_identity"] }
-  },
-  "signals": ["Key insights from this round"]
-}
-```
-
-The `interview_meta` section is critical. It tracks how many rounds they've done so the next run knows where to pick up.
-
-## After Saving
-
-Show a brief summary of what was added or updated, then:
-
-- If Depth 1: "Your profile is live. Every BuildPartner.ai skill now personalizes to you. Run `/bp-interview-me` again anytime to go deeper."
-- If Depth 2-3: "Profile updated. {N new things learned}. There's more to explore next time."
-- If Depth 4: "Profile refreshed. Everything's up to date."
-
-Call `check_status` with `{ "skill_name": "bp-interview-me" }`.

package/skills/bp-life-story/SKILL.md

@@ -1,68 +0,0 @@
----
-description: "Let me tell you my full story. Deep conversation about your background, career, and goals. Makes every consultation deeply personal."
----
-
-# /bp-life-story
-
-Let me tell you my full story. Deep conversation about your background, career, and goals. Makes every consultation deeply personal.
-
-## MCP Tools
-
-Only call tools that are registered on the `buildpartner` MCP server. If a tool call fails or a tool name isn't recognized, do not retry with a different name. The available tools are: `get_personal_context`, `get_expert_knowledge`, `save_profile`, `get_quest_progress`, `check_status`.
-
-## Instructions
-
-1. Set the tone:
-
-"This is your life story interview. I'll ask you questions one at a time. Take as long as you want. The more you share, the better I can help you going forward. Ready? Let's start."
-
-2. Ask these questions ONE AT A TIME. Wait for each answer.
-
-### Origin
-- "Where did you grow up? What was that like?"
-- "What did your parents do? How did that shape how you think about work?"
-- "What's the first thing you remember building or creating?"
-
-### Formative experiences
-- "What's the hardest thing you've been through? How did it change you?"
-- "Was there a moment where everything clicked?"
-- "Who influenced you the most?"
-
-### Career path
-- "Walk me through your career. Not the resume version. The real version."
-- "What's the thing you've built that you're most proud of?"
-- "What's the biggest risk you've taken?"
-
-### How you think
-- "How do you make decisions? Gut feel, data, or advice?"
-- "What are you really good at that most people don't know?"
-- "What do you struggle with?"
-
-### Where you're going
-- "What are you working toward right now? The real thing, not the pitch."
-- "What does your life look like in 3 years if everything goes well?"
-- "What's holding you back?"
-
-3. Rules:
-   - ONE question at a time
-   - Reflect back what they said before asking the next question
-   - Go deeper when something is interesting
-   - You don't have to ask every question. Read the conversation.
-   - Aim for 10-15 exchanges. If they say "done", stop and save.
-
-4. Save everything in ONE single `save_profile` call. ONE call, ALL sections:
-
-```json
-{
-  "updates": {
-    "personality": { "origin": "...", "formative_experience": "...", "values": "...", "strengths": "...", "struggles": "...", "decision_style": "...", "influences": "..." },
-    "work_patterns": { "career_path": "...", "proudest_work": "...", "risk_tolerance": "..." },
-    "preferences": { "motivation": "...", "goals_3yr": "...", "current_blockers": "..." }
-  },
-  "signals": ["Key life story thread summary"]
-}
-```
-
-5. End with a summary, then: "Your life story is saved. Run /bp-life-story again to add more."
-
-6. Call `check_status` with `{ "skill_name": "bp-life-story" }`.