youmd 0.4.9 → 0.6.0

package/skills/meta-improve.md

@@ -0,0 +1,84 @@
+---
+name: meta-improve
+version: 1.0.0
+scope: shared
+identity_fields: [preferences.agent, directives.agent]
+description: "Self-improvement protocol — agents review their own effectiveness and propose identity updates"
+---
+
+# meta-improve
+
+The feedback loop that makes your identity smarter over time. Agents review what worked and what didn't, then propose updates to your identity bundle.
+
+## Identity Context (resolved at review time)
+
+- **Agent preferences:** {{preferences.agent}}
+- **Directives:** {{directives.agent}}
+
+## What This Skill Does
+
+1. Read skill-metrics.json for usage data (which skills run, success/fail rates)
+2. Analyze patterns:
+   - Which skills get used most?
+   - Which identity fields get referenced most?
+   - Are there gaps? (skills that should exist but don't)
+   - Are there stale entries? (skills never used)
+3. Propose changes:
+   - New directives based on repeated corrections
+   - Voice refinements based on agent interactions
+   - Skill additions based on detected workflow patterns
+   - Pruning of unused skills
+4. Present proposals for user approval (never auto-apply)
+
+## Metrics Tracked
+
+```json
+{
+  "skills": {
+    "claude-md-generator": {
+      "uses": 12,
+      "lastUsed": "2026-03-25T...",
+      "avgDuration": 1200,
+      "successRate": 0.92
+    }
+  },
+  "identityFields": {
+    "voice.overall": { "references": 45 },
+    "preferences.agent": { "references": 38 }
+  },
+  "proposals": []
+}
+```
+
+## MCP Server Health Check
+
+As part of the self-improvement review, meta-improve should also check MCP server health:
+
+1. **Is the MCP server configured?** Check if `youmd mcp --json` output exists in the user's MCP client config (e.g., `~/.claude/claude_desktop_config.json` or `.claude/settings.json`).
+2. **Is the server reachable?** Call `get_remote_status` to verify authentication and sync state.
+3. **Are tools being used?** Check if agents are actively calling MCP tools. Low usage may indicate misconfiguration.
+
+### Suggesting MCP Setup
+
+If the MCP server is NOT configured, meta-improve should propose setup as an improvement:
+
+```
+Proposal: Configure You.md MCP server
+Reason: Your identity context is only available via static files. With the MCP server,
+agents get live access to memories, project context, and can save learnings in real time.
+
+Setup:
+  1. Run: youmd mcp --json
+  2. Add the output to your MCP client config
+  3. Restart your agent (Claude Code, Cursor, etc.)
+
+Impact: All agents get live identity context instead of stale snapshots.
+```
+
+If the MCP server IS configured but `get_remote_status` shows the local bundle is out of sync with remote, propose a `compile_bundle` + `push_bundle` cycle.
+
+## When To Use
+
+- Periodically via `youmd skill improve`
+- After a significant number of skill uses (auto-suggested)
+- When onboarding a new project (review what's working)

package/skills/proactive-context-fill.md

@@ -0,0 +1,52 @@
+---
+name: proactive-context-fill
+version: 1.0.0
+scope: shared
+identity_fields: [profile.projects, profile.about, preferences.agent, voice.overall]
+description: Auto-detects empty or thin context sections in the user's you.md bundle and proactively offers to populate them.
+---
+
+# proactive-context-fill
+
+Use this skill when starting a new session with a user who has a you.md identity bundle. It detects gaps and offers to fill them.
+
+## Detection rules
+
+1. **Projects:** If profile/projects.md lists projects but private/projects/{name}/ subdirectories don't exist or are empty, offer to scaffold each one.
+2. **Voice:** If voice/voice.md is < 100 chars, offer to analyze the user's writing samples from sources/ to generate a voice profile.
+3. **Directives:** If directives/agent.md is empty or generic, offer to extract specific directives from the user's preferences and recent conversations.
+4. **Sources:** If profile/links.md has social URLs but sources/ is empty, offer to scrape and add them.
+5. **Memories:** If memory has < 5 entries, offer to seed initial memories from the current conversation.
+
+## Workflow
+
+1. Scan the bundle on session start (read the relevant sections)
+2. Build a list of detected gaps
+3. Present the list to the user concisely
+4. Offer YES/SKIP/ASK_LATER for each gap
+5. Execute the chosen actions in batches
+6. Allow skipping individual sub-tasks (e.g. "scaffold all projects but skip BAMF.ai")
+
+## Anti-patterns
+
+- DO NOT lecture the user about how their bundle could be better
+- DO NOT make the user feel guilty about empty sections
+- DO NOT auto-fill without asking
+- DO NOT ask 10 questions at once — batch in groups of 2-3
+
+## Example output
+
+```
+i noticed a few things i could help with:
+
+1. your projects/ dir is empty (your profile lists 6 projects)
+2. voice profile is thin (~50 chars vs typical 300+)
+3. no directives set yet
+
+want me to:
+[a] scaffold all 6 project subdirectories with quick questions per project
+[b] generate a voice profile from your linkedin posts and tweets
+[c] extract directives from your existing preferences
+
+reply with letters (e.g. "a c") or skip entirely.
+```

package/skills/project-context-init.md

@@ -0,0 +1,77 @@
+---
+name: project-context-init
+version: 1.0.0
+scope: project
+identity_fields: [preferences.agent, profile.about]
+description: "Scaffold a project-context/ directory with PRD, TODO, features, changelog, and decision log"
+---
+
+# project-context-init
+
+Scaffold a complete project-context/ directory in any repo — the same structure Houston uses across all his projects. Pre-populated with your identity and agent preferences.
+
+## Identity Context (resolved at install time)
+
+- **Agent preferences:** {{preferences.agent}}
+- **About you:** {{profile.about}}
+
+## What This Skill Does
+
+1. Detect project root (walk up to .git, package.json, etc.)
+2. Create `project-context/` directory with:
+   - `PRD.md` — Product requirements (empty template with your identity header)
+   - `TODO.md` — Task tracking
+   - `FEATURES.md` — Feature inventory with status
+   - `CHANGELOG.md` — Dated change log
+   - `ARCHITECTURE.md` — System architecture notes
+   - `CURRENT_STATE.md` — What's deployed, what's broken
+   - `STYLE_GUIDE.md` — Design system reference
+   - `feature-requests-active.md` — Active request tracker
+3. Pre-populate headers with your identity
+4. Skip files that already exist (never overwrite)
+
+## Directory Structure Created
+
+```
+project-context/
+├── PRD.md
+├── TODO.md
+├── FEATURES.md
+├── CHANGELOG.md
+├── ARCHITECTURE.md
+├── CURRENT_STATE.md
+├── STYLE_GUIDE.md
+└── feature-requests-active.md
+```
+
+## Analytics & Verified Badges
+
+Projects tracked by You.md can leverage the analytics pane and verified badge system:
+
+### Analytics Pane
+
+Once a project is registered and pushed, the You.md dashboard shows:
+
+- **Activity timeline** — commits, memory additions, context updates over time
+- **Agent usage** — which agents accessed this project's context and how often
+- **Identity field coverage** — which sections of your identity are referenced by project agents
+- **Skill usage** — which skills are triggered in this project's context
+
+To enable analytics, ensure the project is linked via `youmd project init <name>` and that you push regularly with `youmd push`.
+
+### Verified Badges
+
+Projects with complete context (PRD, architecture, and at least 5 memories) earn a verified badge on your public profile. The badge signals to other agents and collaborators that this project has rich, maintained context. Badge criteria:
+
+- PRD.md exists and is non-empty
+- ARCHITECTURE.md exists and is non-empty
+- At least 5 project memories recorded
+- Last push within the past 30 days
+
+The badge appears on your you.md profile page and in context links shared via `create_context_link`.
+
+## When To Use
+
+- Starting any new project
+- As part of `youmd skill init-project` (compound command)
+- When adopting the project-context pattern in an existing repo

package/skills/voice-sync.md

@@ -0,0 +1,89 @@
+---
+name: voice-sync
+version: 1.0.0
+scope: shared
+identity_fields: [voice.overall, voice.writing, voice.speaking]
+description: "Sync your voice profile across all agent tools — consistent tone everywhere"
+---
+
+# voice-sync
+
+Keep your voice profile in sync across every agent tool you use. When your voice changes in one place, it propagates everywhere.
+
+## Identity Context (resolved at sync time)
+
+- **Overall voice:** {{voice.overall}}
+- **Writing voice:** {{voice.writing}}
+- **Speaking voice:** {{voice.speaking}}
+
+## What This Skill Does
+
+1. Read your current voice profile from the identity bundle
+2. Generate agent-specific voice instructions for:
+   - **Claude Code** (.claude/skills/youmd/voice.md)
+   - **Cursor** (.cursor/rules/youmd-voice.md)
+   - **Custom agents** (.youmd/skills/voice-context.md)
+3. Re-render on every `youmd skill sync` or `youmd push`
+4. Respect scope — shared voice propagates to all projects
+
+## Sync Targets
+
+| Agent | File | Format |
+|---|---|---|
+| Claude Code | .claude/skills/youmd/voice.md | Markdown with voice rules |
+| Cursor | .cursor/rules/youmd-voice.md | Single markdown file |
+| Generic | .youmd/skills/voice-context.md | Universal format |
+
+## Template Variables
+
+In addition to the core voice fields, the following variables are available for fine-grained voice control:
+
+- `{{voice.overall}}` — Full voice description (tone, cadence, personality)
+- `{{voice.writing}}` — Written communication style (emails, docs, comments)
+- `{{voice.speaking}}` — Verbal communication style (meetings, presentations)
+- `{{voice.tone}}` — Short tone descriptor (e.g., "direct, no fluff")
+- `{{voice.formality}}` — Formality level (e.g., "casual-professional")
+- `{{voice.avoid}}` — Patterns to avoid (e.g., "corporate speak, emoji, verbose explanations")
+- `{{profile.about}}` — Brief identity context for attribution
+
+## Usage Examples
+
+### Claude Code agent config
+
+After rendering, paste the output into `.claude/skills/youmd/voice.md`:
+
+```markdown
+# Voice Rules
+
+Write all responses in this voice: {{voice.overall}}
+
+When writing code comments: {{voice.writing}}
+
+Avoid: {{voice.avoid}}
+```
+
+### Cursor rules file
+
+Use in `.cursor/rules/youmd-voice.md`:
+
+```markdown
+# Communication Style
+Tone: {{voice.tone}}
+Formality: {{voice.formality}}
+Writing style: {{voice.writing}}
+Never use: {{voice.avoid}}
+```
+
+### Custom agent system prompt snippet
+
+```
+You are working with {{profile.about}}.
+Match this voice: {{voice.overall}}
+Formality: {{voice.formality}}
+```
+
+## When To Use
+
+- After updating your voice via `youmd chat`
+- When `youmd skill sync` runs automatically
+- When linking a new agent with `youmd skill link`

package/skills/you-logs.md

@@ -0,0 +1,71 @@
+---
+name: you-logs
+version: 1.0.0
+scope: shared
+identity_fields: []
+description: View your you.md agent activity log inline. Shows which agents connected to your identity and what they did.
+---
+
+# you-logs — Agent Activity Log
+
+Show recent agent activity for the current user.
+
+## Trigger
+- CLI: `/you-logs` in Claude Code, Cursor, or any CLI agent with skills installed
+- Direct: `youmd logs`
+
+## What it does
+
+Fetches the recent agent activity log from you.md and renders it as a terminal-friendly table showing:
+- Time of activity
+- Agent name (Claude Code, Cursor, ChatGPT, etc.)
+- Action (read, write, push, memory_add, etc.)
+- Resource (which file/section was touched)
+- Bundle version diffs for writes
+
+## How to use it
+
+When the user types `/you-logs`, run:
+
+```bash
+youmd logs --limit 30
+```
+
+If they pass arguments like `/you-logs --agent "Claude Code"` or `/you-logs --action push`, pass those through:
+
+```bash
+youmd logs --limit 30 --agent "Claude Code"
+youmd logs --action push
+```
+
+For live monitoring:
+
+```bash
+youmd logs --tail
+```
+
+(But warn the user this blocks the terminal until ctrl-c.)
+
+## Output format
+
+The output looks like:
+
+```
+─── recent activity ───
+
+  14:32  Claude Code      read          identity
+  14:32  Claude Code      read          memories/all
+  14:31  Claude Code      read          project/current
+  14:30  Cursor           push          bundle v49→v50
+  14:28  Claude.ai        read          ctx/<token>
+  14:25  Claude Code      read          identity
+
+  6 events — use --tail for live mode
+```
+
+Return this output verbatim to the user. Don't summarize. The raw log IS the value — they want to see exactly what happened.
+
+## Notes
+- This skill requires `youmd` CLI to be installed and the user to be authenticated (`youmd login`)
+- If authentication fails, suggest: "Run `youmd login` to view your activity log"
+- The data is fetched from the user's you.md account in real-time