Skip to content

moodler/liz-tracker

BranchesTags

Repository files navigation

Liz Tracker

Liz Tracker

Run your life efficiently from any device

A self-hosted project management tracker with a beautiful kanban UI, purpose-built workspaces, AI agent integration, and an optional orchestrator that automatically dispatches work to AI coding sessions.

Created by Martin Dougiamas. I am building this for myself, and development is VERY active. I'm happy to take requests but I'm more focussed on my needs than yours right now! That said, I'm trying to keep it useful for anyone.

What You Get

Kanban Dashboard

A dark-themed, drag-and-drop kanban board that works beautifully on desktop, tablet, and phone. Each platform gets its own optimised layout — not just a responsive resize, but a genuinely different touch-friendly UI on mobile. Installable as a PWA so it feels like a native app.

Spaces — Purpose-Built Workspaces

Work items aren't just tickets. Each item can open into a full-screen workspace tailored to its type:

Space What it does
Standard Classic tracker item — detail panel with description, comments, attachments
Song Songwriting workspace — split-pane lyrics editor with version history, conversation sidebar, metadata bar (genre, key, BPM), cover art, and inline CriticMarkup comments
Text Writing workspace — markdown editor with conversation sidebar and inline CriticMarkup comments for articles, blogs, or any long-form text
Engagement Coordination hub for contractors and services — contact card, quote/financials, milestone timeline, document attachments, communications log, and discussion sidebar
Scheduled Recurring task manager — schedule config (daily/weekly/custom cron), live status panel, task instructions, TODO and IGNORE lists, run history
Travel Trip planner — day-by-day itinerary with timezone-aware segments (flights, lodging, transport, activities, restaurants, meetings), automatic gap detection, and cover images
Presentation Presentation development workspace — 3-tab layout (Description, Slides, Deck) with discussion sidebar and DeckWright integration

Projects choose which space types are available, and you pick the type when creating an item.

Work Item Pipeline

Items flow through a clear lifecycle:

brainstorming → clarification → approved → in_development → in_review → testing → done
  • Brainstorming — capture ideas and define requirements
  • Clarification — optionally trigger a research agent to gather info and flesh out the spec
  • Approved — human-approved and ready for work (or automatic dispatch to AI)
  • In development → In review → Testing → Done — standard workflow
  • Needs input — blocked, waiting for human input
  • Cancelled — archive from any state

AI Agent Integration

Any MCP-compatible AI agent can manage your tracker programmatically via 65+ tools — create items, update status, add comments, propose batched changes, find semantically similar items, manage travel segments, update engagement milestones, and more. Connect your agent to the MCP endpoint and it can work alongside you.

AI Orchestrator

Optionally let the tracker automatically dispatch approved work items to AI coding sessions. Two dispatch backends are supported:

  • Session Runner (DISPATCH_MODE=runner) — runs Claude Code directly via the Agent SDK as a child process. No external services needed. Includes a dashboard session viewer with live activity feed, human steering, and persistent transcripts.
  • OpenCode (DISPATCH_MODE=opencode) — dispatches to OpenCode sessions via its SDK.

The orchestrator:

  • Polls for approved items and creates coding sessions automatically
  • Sends research agents to flesh out specs during the clarification phase
  • Monitors progress via SSE and updates item status in real time
  • Comment-based auto-completion — when the owner comments "looks good", "done", "LGTM" etc. on items in testing or in_review, the orchestrator auto-advances them to done
  • Review feedback redispatch — when the owner leaves non-acknowledgment feedback on items in testing, the orchestrator automatically moves them back to in_review and dispatches a new coder session to address the feedback
  • Scheduled task time gating — scheduled tasks wait for their configured time/day before dispatching, with timezone-aware checks for daily, weekly, hourly, and monthly frequencies
  • Per-task model selection — scheduled tasks can specify a model strength tier (high/medium/low) to control which Claude model is used for execution, instead of always using the global default
  • Recurring scheduled task recycling — when a recurring scheduled task completes, the orchestrator automatically recycles it back to approved for the next execution cycle, preserving original human approval provenance
  • Expired scheduled task auto-close — scheduled tasks with a past due date are automatically moved to done
  • Dashboard session viewer — watch agent activity in real-time (tool calls, reasoning, errors), send steering messages to redirect the agent, and review session transcripts after completion
  • Includes safety features: actor classification (only humans can approve), description integrity checks, circuit breakers, per-item retry limits, and an emergency stop button

Other Features

  • Comments and discussion — threaded comments on any item, with emoji reactions and inline CriticMarkup in Song and Text spaces
  • Attachments — upload files and images to any item, including paste-from-clipboard support
  • Dependencies — link items that block each other
  • Typed links — relate items with typed relations (relates_to, duplicates, parent_of, etc.); KEY-N mentions in titles and descriptions auto-create links
  • Groups — turn any item into a group by adding parent_of links to other items; groups inherit state/comments/search for free, support drag-reorder, multi-select "Group as new item", cross-project membership, and kanban progress mini-bars ("12/15 done")
  • Refactor primitives — merge multiple items into one, split one item into many, or bulk-edit labels/priority/state across a selection in a single transaction (with version snapshots for reversibility)
  • Semantic discovery — embeddings power a Smart Related panel that surfaces similar items, plus optional Merge Candidates and Topics views; suggestions can be confirmed or dismissed (dismissals remembered so they don't return)
  • Proposals — agents stage batches of multi-action suggestions (merges, splits, bulk updates, etc.) for human review; nothing executes until a human applies them
  • Cover images — visual cover art for Song and Travel items
  • Deep links — shareable URLs that open directly to any item, with Open Graph meta tags for rich link previews in iMessage, Slack, etc.
  • Activity log — unified timeline of all mutations (state changes, comments, description edits, attachments) with actor attribution and filtering
  • AI categorization — one-click AI-powered field extraction from description text
  • Search — full-text search across all items
  • Today dashboard — cross-project attention view showing items needing action, with priority sorting and due date display
  • Project reordering — drag-and-drop project tab ordering
  • Webhook notifications — optional comment webhook for external integrations
  • Settings dialog — configure the default coder model, effort level, and scheduled-task strength tiers from the dashboard topbar (dynamic model list fetched from Anthropic API)
  • LAN access — accessible from any device on your network, installable as a PWA

Quick Start

# Install dependencies
npm install

# Copy and edit configuration
cp .env.example .env

# Run in development mode (hot reload)
npm run dev

# Or build and run
npm run build
npm start

The server starts on port 1000 by default. Open http://localhost:1000 for the dashboard.

Configuration

Copy .env.example to .env and edit as needed. Key settings:

Variable Default Description
PORT 1000 Server port
STORE_DIR ./store Directory for SQLite database and attachments
OWNER_NAME Owner Your display name (used as default assignee)
HUMAN_ACTORS dashboard,me Comma-separated names classified as human (can approve items)
AGENT_ACTORS coder Comma-separated names classified as AI agents
TRACKER_API_TOKEN (auto-generated) Bearer token for API endpoints
TRACKER_PUBLIC_URL http://localhost:{PORT} Tracker dashboard URL (for links and fallback base)
TRACKER_SHORT_URL (same as TRACKER_PUBLIC_URL) Short base URL for item deep links (e.g. http://t → http://t/TRACK-187)
ANTHROPIC_API_KEY (none) Enables AI categorization in the dashboard
AI_CATEGORIZE_MODEL claude-haiku-4-5-20251001 Model for AI categorization
WEBHOOK_URL (none) URL to POST comment webhook notifications to
WEBHOOK_SECRET (none) Shared secret for authenticating webhook payloads
ASSISTANT_PROJECT_ROOT ~/assistant Root directory for container path translation (agent file uploads)

On first run, an API token is auto-generated and saved to store/auth_token. See .env.example for all options including orchestrator settings.

Architecture

  • Runtime: Node.js + TypeScript, vanilla JS frontend (no frameworks)
  • Database: SQLite with WAL mode — single file, no external dependencies
  • UI: Pre-compiled from src/ui/core.html + src/ui/spaces/*.js into a single index.html
  • API: REST at /api/v1/ + MCP at /mcp (Streamable HTTP, stateless)
src/
├── index.ts          # Entry point
├── config.ts         # Environment configuration
├── db.ts             # SQLite database layer (schema, CRUD, migrations, activity log, links, proposals, embeddings)
├── api.ts            # HTTP server — REST API + static files + MCP routing + OG meta tag injection
├── mcp-server.ts     # MCP tool definitions (65+ tools, including dynamic space plugin tools)
├── orchestrator.ts   # AI orchestrator — dispatches work to coding sessions, monitors progress
├── session-runner.ts # Session runner — direct Claude Code execution via Agent SDK
├── runner-types.ts   # Shared types for runner stdio JSON protocol
├── runner-output.ts  # Runner output helpers (truncation, arg summaries, unified diffs)
├── embeddings.ts     # Embedding provider abstraction (Voyage, local/mock); vector math
├── embeddings-worker.ts # Debounced refresh queue, nightly neighbour/drift/clustering job
├── logger.ts         # Pino logger
├── spaces/           # Space plugin backends (types, registry, per-space logic)
│   ├── types.ts      # SpacePlugin interface
│   ├── registry.ts   # Plugin registration
│   ├── index.ts      # Registration manifest
│   ├── standard.ts   # Standard space (identity only)
│   ├── song.ts       # Song space backend
│   ├── text.ts       # Text space backend
│   ├── engagement.ts # Engagement space backend (6 routes, 7 MCP tools)
│   ├── scheduled.ts  # Scheduled space backend (4 routes, 4 MCP tools)
│   ├── travel.ts     # Travel space backend (4 routes, 4 MCP tools)
│   └── presentation.ts # Presentation space backend (4 routes, DeckWright integration + thumbnail proxy)
└── ui/
    ├── core.html     # Dashboard shell + plugin registry + overlay
    └── spaces/       # Per-space UI renderers (JS plugins)

Adding a New Space

The space system is fully pluggable — add a new space type with just 2 files + 1 registry line:

  1. Create src/spaces/{name}.ts — backend (parser, API routes, MCP tools)
  2. Create src/ui/spaces/{name}.js — frontend renderer
  3. Add registerSpace({name}Plugin) to src/spaces/index.ts
  4. Run npm run build

No changes needed to api.ts, mcp-server.ts, db.ts, or the UI shell.

REST API

Write endpoints require Authorization: Bearer <token>. Read endpoints are unauthenticated.

Projects

Method Path Description
GET /api/v1/projects List all projects
POST /api/v1/projects Create a project
PUT /api/v1/projects/reorder Reorder project tabs
GET /api/v1/projects/:id Get a project
PATCH /api/v1/projects/:id Update a project (name, description, context, theme, working_directory, opencode_project_id, orchestration, active_spaces)
DELETE /api/v1/projects/:id Delete a project
GET /api/v1/projects/:id/stats Get project statistics
GET /api/v1/projects/:id/tracker Get kanban-grouped view for a project

Work Items

Method Path Description
GET /api/v1/items List items (filterable by project, state, priority, assignee, search)
POST /api/v1/items Create an item
GET /api/v1/items/:id Get an item (with comments, transitions, dependencies, attachments)
PATCH /api/v1/items/:id Update an item (supports space_type, space_data, project_id for cross-project moves)
DELETE /api/v1/items/:id Delete an item
POST /api/v1/items/:id/state Transition item state
POST /api/v1/items/:id/lock Lock an item
POST /api/v1/items/:id/unlock Unlock an item
POST /api/v1/items/clear-stale-locks Clear locks older than 2 hours
GET /api/v1/items/recent Recently updated items (filterable by project, with limit)
POST /api/v1/items/ai-categorize AI-powered field extraction from description text

Comments, Dependencies, Attachments

Method Path Description
GET /api/v1/items/:id/comments List comments
POST /api/v1/items/:id/comments Add a comment
PATCH /api/v1/comments/:id Update a comment
DELETE /api/v1/comments/:id Delete a comment
POST /api/v1/comments/:id/reactions Toggle an emoji reaction on a comment
GET /api/v1/comments/:id/reactions Get aggregated reactions for a comment
GET /api/v1/items/:id/dependencies List dependencies
POST /api/v1/items/:id/dependencies Add a dependency
DELETE /api/v1/items/:id/dependencies/:depId Remove a dependency
GET /api/v1/items/:id/attachments List attachments
POST /api/v1/items/:id/attachments Upload an attachment (JSON with base64 or multipart/form-data)
GET /api/v1/attachments/:id Serve attachment file
GET /api/v1/attachments/:id/meta Get attachment metadata
DELETE /api/v1/attachments/:id Delete an attachment
GET /api/v1/items/:id/transitions List state transitions
GET /api/v1/items/:id/watchers List watchers
POST /api/v1/items/:id/watchers Add a watcher
DELETE /api/v1/items/:id/watchers/:entity Remove a watcher

Typed Links and Groups

Method Path Description
GET /api/v1/items/:id/links List typed links for an item (query: ?relation=)
POST /api/v1/items/:id/links Add a typed link ({to_item_id, relation, note?})
DELETE /api/v1/items/:id/links/:linkId Remove a typed link
POST /api/v1/items/:id/links/:linkId/confirm Promote an embedding-suggested link to a manual link
GET /api/v1/items/:id/children List children + parent (for items with parent_of links)
PATCH /api/v1/items/:id/children/reorder Reorder children ({child_ids: [...]})
POST /api/v1/items/group Create a new group from selected items ({title, child_item_ids, description?, target_project_id?})

Refactor and Bulk Operations

Method Path Description
POST /api/v1/items/merge Merge N source items into a target (single transaction, reversible)
POST /api/v1/items/:id/split Split one item into N children with optional regex-based comment extraction
PATCH /api/v1/items/bulk Bulk-update many items in one transaction (labels, priority, assignee, state, project move, add_links)

Embeddings and Semantic Discovery

Method Path Description
GET /api/v1/items/:id/similar Top-K semantically similar items + drift score
GET /api/v1/embeddings/status Aggregate stats (counts, provider, thresholds, last neighbour run)
POST /api/v1/embeddings/recompute Re-enqueue items for embedding (admin)
GET /api/v1/embeddings/merge-candidates Global high-similarity pairs
GET /api/v1/embeddings/topics All clusters with members
PATCH /api/v1/embeddings/clusters/:id/label Rename a cluster
POST /api/v1/embeddings/tombstones Record "not duplicates" so the pair won't be re-proposed
GET /api/v1/embeddings/tombstones List all tombstones

Proposals

Method Path Description
POST /api/v1/proposals Stage a new proposal ({title, summary?, expires_in_days?, actions: [...]})
GET /api/v1/proposals List proposals (?status=&since=&limit=)
GET /api/v1/proposals/:id Get a proposal with all actions
PATCH /api/v1/proposals/:id/actions/:action_id Accept or reject a single action
POST /api/v1/proposals/:id/apply Apply accepted actions (human-class actors only)
DELETE /api/v1/proposals/:id Reject the entire proposal

Space-Specific Endpoints

Method Path Description
GET /api/v1/items/:id/versions Description version history
POST /api/v1/items/:id/versions Save a description version snapshot
POST /api/v1/items/:id/versions/revert Revert description to a specific version
PUT /api/v1/items/:id/cover Set/replace cover image
DELETE /api/v1/items/:id/cover Remove cover image
POST /api/v1/items/:id/scheduled/todo Add TODO items
DELETE /api/v1/items/:id/scheduled/todo Remove TODO items
POST /api/v1/items/:id/scheduled/ignore Add IGNORE rules
DELETE /api/v1/items/:id/scheduled/ignore Remove IGNORE rules
PATCH /api/v1/items/:id/engagement/contact Update engagement contact
PATCH /api/v1/items/:id/engagement/quote Update engagement quote/financials
POST /api/v1/items/:id/engagement/milestones Add milestones
DELETE /api/v1/items/:id/engagement/milestones Remove milestones
POST /api/v1/items/:id/engagement/comms Add comms log entries
PATCH /api/v1/items/:id/engagement/settings Update engagement settings
PATCH /api/v1/items/:id/travel/trip Update travel trip metadata
POST /api/v1/items/:id/travel/segments Add travel segments
PATCH /api/v1/items/:id/travel/segments Update a travel segment
DELETE /api/v1/items/:id/travel/segments Remove travel segments
PATCH /api/v1/items/:id/presentation/deck Update deck config (slug + URL)
GET /api/v1/items/:id/presentation/deck-mdx Read deck MDX from DeckWright
GET /api/v1/items/:id/presentation/deck-thumbnails Fetch slide thumbnails (cached proxy, ?refresh=1 to bust cache)
GET /api/v1/items/:id/presentation/deck-thumb Serve a cached slide thumbnail image

Activity Log

Method Path Description
GET /api/v1/activity List recent activity (global, filterable by project, item, action, actor, since, search)
GET /api/v1/projects/:id/activity Activity for a specific project
GET /api/v1/items/:id/activity Activity for a specific item

Cross-Project Views

Method Path Description
GET /api/v1/attention Items needing attention (cross-project)
GET /api/v1/overview Cross-project kanban view with all items
GET /api/v1/search?q=... Full-text search across items (supports issue key lookup)

Orchestrator

Method Path Description
GET /api/v1/orchestrator/status Get orchestrator status
POST /api/v1/orchestrator/pause Pause the orchestrator
POST /api/v1/orchestrator/resume Resume the orchestrator
POST /api/v1/orchestrator/emergency-stop Emergency stop
POST /api/v1/orchestrator/restart Safe restart
GET /api/v1/orchestrator/restart Check restart status
DELETE /api/v1/orchestrator/restart Cancel pending restart
GET /api/v1/orchestrator/safe-to-restart Check if safe to restart
POST /api/v1/items/:id/dispatch Manually dispatch an item
GET /api/v1/items/:id/session Get session info for an item
POST /api/v1/items/:id/session/abort Abort active session for an item
GET /api/v1/items/:id/session/events SSE stream of session events (runner mode)
POST /api/v1/items/:id/session/steer Send steering message to running agent (runner mode)
GET /api/v1/items/:id/audits Execution audits for an item (includes transcripts)

Utility

Method Path Description
GET /api/v1/states List valid states and priorities
GET /api/v1/config Public dashboard configuration
GET /api/v1/settings Tracker-wide settings (coder model, effort, strength-tier models)
PATCH /api/v1/settings Update tracker-wide settings
GET /api/v1/models List available Anthropic models (requires ANTHROPIC_API_KEY)
POST /api/v1/auth/verify Verify API token
GET /api/v1/auth/status Check authentication status

MCP Server

The MCP endpoint at /mcp (Streamable HTTP, stateless) exposes 65+ tools for AI agents. Connect any MCP-compatible client to http://localhost:1000/mcp.

Tools cover: project and item CRUD, state transitions, comments, comment reactions, watchers, dependencies, typed links, groups (list children, create group), refactor primitives (merge, split, bulk update), batch proposals (propose, apply, list, get), semantic similarity (find similar), attachments, orchestrator control (dispatch, abort, emergency stop, safe restart), cover images, agent config validation, agent reference documentation, and all space-specific operations (scheduled TODOs/IGNORE rules, engagement milestones/contacts/quotes/comms/settings, travel segments/trips).

AI Orchestrator Setup

The orchestrator is disabled by default. Two dispatch backends are available.

Option A: Session Runner (recommended)

Uses the Claude Agent SDK to run Claude Code directly — no external services needed.

  1. Enable in .env:

    ORCHESTRATOR_ENABLED=true
DISPATCH_MODE=runner

  2. Set a working directory on each project you want to orchestrate (via project settings in the dashboard).

  3. Ensure Claude Code is installed and authenticated (the runner uses your Claude Max subscription or API key).

That's it. The tracker spawns Claude Code sessions as child processes and communicates via stdio JSON.

Option B: OpenCode

Uses OpenCode as the session manager.

  1. Install OpenCode and start the server:

    opencode serve --hostname 0.0.0.0 --port 3000

  2. Connect OpenCode to the tracker's MCP server — add to your OpenCode config:

    {
  "mcp_servers": {
    "tracker": {
      "type": "http",
      "url": "http://localhost:1000/mcp",
      "enabled": true
    }
  }
}

  3. Enable in .env:

    ORCHESTRATOR_ENABLED=true
DISPATCH_MODE=opencode
OPENCODE_SERVER_URL=http://localhost:3000

  4. Set a working directory on each project you want to orchestrate (via project settings in the dashboard).

Orchestrator Configuration

Variable Default Description
ORCHESTRATOR_ENABLED false Master switch
DISPATCH_MODE opencode runner (direct Claude Code) or opencode (OpenCode SDK)
OPENCODE_SERVER_URL http://localhost:3000 OpenCode server URL for API calls
OPENCODE_PUBLIC_URL (same as server URL) OpenCode URL for browser deep links
ORCHESTRATOR_INTERVAL 30000 Poll interval (ms)
OPENCODE_MAX_CONCURRENT 3 Max concurrent sessions globally
OPENCODE_MAX_PER_PROJECT 1 Max concurrent sessions per project
SESSION_TIMEOUT 2700000 Session timeout (ms, default 45 minutes)
CODER_MODEL_PROVIDER anthropic Model provider for coder sessions
CODER_MODEL_ID claude-opus-4-6 Model ID for coder sessions
CODER_EFFORT high Default reasoning effort (low/medium/high/max)
CIRCUIT_BREAKER_THRESHOLD 2 Consecutive failures before auto-pause
CIRCUIT_BREAKER_WINDOW 3600000 Failure counting window (ms, default 1 hour)
ITEM_DISPATCH_FAILURE_LIMIT 3 Per-item failures before auto-shelving

All model/effort defaults above can be overridden at runtime via the dashboard's Settings dialog (topbar gear icon). Stored overrides live in the tracker_settings DB table and take precedence over env-var defaults.

Safety Features

  • Actor classification — only human actors can approve items for execution
  • Description integrity — SHA-256 hash verified at dispatch to detect tampering
  • Circuit breaker — auto-pauses after consecutive failures
  • Per-item retry limit — auto-shelves items that fail repeatedly
  • Emergency stop — dashboard button to pause orchestrator and cancel all sessions
  • Safe restart — waits for active sessions to complete before restarting
  • Blocked file patterns — prevents AI agents from modifying security-critical files
  • Execution audits — full audit trail of every dispatch (item, session, prompt hash, status)
  • Session timeout — stale sessions auto-detected and aborted (default 45 minutes)
  • Agent config validation — pre-flight check before dispatch to ensure agent config is valid
  • Session recovery — on tracker restart, recovers active sessions and polls their status
  • Scheduled task recycling — recurring scheduled tasks are automatically recycled back to approved after completion, preserving approval provenance
  • Scheduled task time gating — tasks only dispatch when their configured schedule time arrives (timezone-aware)

Claude Code Skills, Commands, and Hooks

The project includes custom Claude Code skills, commands, and hooks in .claude/ that guide both interactive developer sessions and orchestrator-dispatched autonomous sessions.

Commands

Slash commands available during Claude Code sessions:

Command Description
/verify Comprehensive pre-commit gate — runs npm run build, npm test, checks for stray console.log, and shows git status
/code-review Security + quality review of uncommitted changes — checks actor classification, esc() usage, blocked file paths, and common issues
/build-fix Incremental TypeScript error fixer — runs tsc, identifies errors, and fixes them with minimal diffs while respecting security-critical file guardrails

Skills

Skills in .claude/skills/ provide domain-specific guidance. They activate automatically based on the work being done, and the orchestrator recommends relevant skills in dispatch prompts based on item content keywords.

Skill Origin Description
tdd-workflow Adapted from ECC Test-driven development discipline for Vitest + in-memory SQLite
security-review Adapted from ECC General security checklist — secrets, SQL injection, XSS, auth patterns
search-first Adapted from ECC Research-before-coding workflow, referencing shared helpers and constants
node-sqlite-patterns Adapted from ECC better-sqlite3 patterns — query optimization, schema design, WAL mode
tracker-security-review Custom Project-specific security — actor classification, approval provenance, description integrity, blocked paths, MCP tool authorization
space-plugin-dev Custom Step-by-step guide for building new space plugins (all 5 parts: backend, frontend, registry, MCP tools, tests)
mcp-tool-dev Custom MCP tool development guide — Zod validation, actor handling, naming conventions, error responses
orchestrator-safe-dev Custom Safety guidelines for orchestrator code — state machine, dispatch, SSE, circuit breaker, safe restart

Hooks

Project-level hooks in .claude/settings.json:

Hook Trigger Description
TypeScript type-check PostToolUse:Edit Runs tsc --noEmit after .ts file edits to catch type errors immediately
Console.log detection Stop Warns about console.log found in modified .ts files (should use logger instead)

Dispatch Integration

When the orchestrator dispatches an item to a coding session, buildPrompt() scans the item's title and description for keywords and injects a "Recommended skills" section into the prompt. This directs the agent to read the relevant .claude/skills/ files before making changes. Keyword categories:

  • tracker-security-review — security, auth, actor, approval, provenance, token, credential, etc.
  • orchestrator-safe-dev — orchestrator, dispatch, session runner, state transition, circuit breaker, SSE, etc.
  • space-plugin-dev — space plugin, new space, space type, registerSpace, etc.
  • mcp-tool-dev — MCP tool, MCP server, new tool, add tool, etc.

Multiple skills can be recommended simultaneously when an item matches multiple categories.

Sub-agent Optimization

Claude Code's Agent tool spawns sub-agents that benefit from Anthropic's prompt caching — the shared system prompt prefix is cached, making sub-agent input tokens ~90% cheaper. This is handled automatically by the claude_code system prompt preset used in both interactive and dispatched sessions. CLAUDE.md includes guidance on when sub-agents are most beneficial for this codebase (parallel exploration, research tasks, independent verifications, multi-file impact analysis).

Testing

npm test              # Run all tests
npm run test:watch    # Watch mode
npm run test:coverage # Coverage report

Tests use Vitest with an in-memory SQLite database. Test suites cover:

  • src/db.test.ts — actor classification, state transitions, security rules, project/item CRUD, locks, dependencies, comments, comment reactions, approval provenance, cross-project moves, activity log
  • src/orchestrator.test.ts — PID-based stale session detection, agent config validation, URL helpers, error classification, scheduled task time gating, per-task model resolution, prompt splitting
  • src/session-runner.test.ts — SDK message mapping, stdio protocol integration tests (event flow, steering)
  • src/spaces/travel.test.ts — type-aware segment deduplication keys

License

GPLv3

About

A self-hosted tracker for managing your life

Resources

Readme

License

GPL-3.0 license
Activity

Stars

11 stars

Watchers

1 watching

Forks

2 forks
Report repository

Releases 1

v1.1.0 — Spaces, Today Dashboard, Scheduled Tasks Latest
Mar 13, 2026

Packages

 
 
 

Contributors

Languages