release

dotclaude

My working ~/.claude/ configuration for Claude Code. Published for reference — not designed as a drop-in clone.

What I'm Exploring

Sessions are ephemeral but work is not. Much of this config is dedicated to an ongoing experiment: what happens when you treat each session as worth remembering?

Chronicle keeps a persistent journal of each session. A hook extracts a memory block at session end (what was accomplished, what's pending, key decisions). At session start, another hook injects relevant context: recent work in this project, pending items, stale threads. Over time: /chronicle curate to organize, /chronicle insights for cross-session patterns, /chronicle publish for digests.

Supporting pieces: a stop hook names each session from its content; analyze-usage runs DuckDB across Claude Code and Cursor logs; recall/remember agents persist memory across sessions; /fork carries context into a new worktree or session.

Much of this config observes work rather than does it. The memory system is where the config is still evolving.

Drawing from This

This config is personal — CLAUDE.md has my name, hooks call my Chronicle scripts, the MCP server needs my API key. To draw from it, cherry-pick rather than clone.

Copy directly (self-contained):

• settings.json permissions pattern (allow/ask/deny tiers)
• skills/status-line-live/ (needs jq + bc)
• Individual commands or skills (each is a standalone directory)

Customize first:

• CLAUDE.md — personal identity and tool preferences
• settings.json hooks — these call Chronicle scripts; remove or replace with your own session lifecycle
• .mcp.json — requires PERPLEXITY_API_KEY environment variable

Prerequisites for the full config:

• bun — hooks and scripts are TypeScript
• jq, bc — statusline calculations
• git — core operations
• Optional: uv (Python), mise (runtimes), Perplexity API key

Project-level .claude/ directories override global settings. See Claude Code docs for the merge behavior.

Directory Structure

~/.claude/
├── CLAUDE.md          # Personal context (name, preferences, tool choices)
├── settings.json      # Permissions, hooks, model selection
├── .mcp.json          # MCP server configs
├── commands/          # Slash commands (/bootstrap, /status_line)
├── skills/            # Extended capabilities with references
├── agents/            # Specialized autonomous agents
├── hooks/             # Session lifecycle scripts
├── scripts/           # Helper utilities
├── references/        # Reference documentation
├── chronicle/         # Chronicle session memory
├── webui/             # Config Visualizer dashboard
├── docs/              # Architecture documentation
└── [session data]     # Gitignored: history, todos, plans, caches, etc.

Quick Reference

Commands (/command)

Command	Purpose
/bootstrap	Scaffold new projects with structure
/claude-webui	Launch the web dashboard for this dotclaude repo
/opensource-precheck	Audit repo before making public
/plan_retro	Append retrospective to current plan
/respond-to-pr-review	Respond to feedback on our PR toward merge
/review-pr	Review a PR someone else wrote
/project-health	Review ~/code projects health
/reflect	Pause to review work with fresh eyes
/reflection-log	Log feedback to improve reflection prompt
/retro	Review session trajectory and update todos
/chronicle wrapup	Deliberate session close-out — curator + conditional backlog update
/chronicle recap	Multi-session narrative recap for a project
/status_line	Explain current session metrics
/update-dependencies	Intelligent dependency updates with batching
/code-review	Review recent work against plan and standards
/codex-review	Code review via non-Claude models

Skills (auto-invoked)