All epics (large, multi-task projects).
Task web UI improvements
Improve the task web UI and pipeline communication: - t-634: Pipeline comment-based agent-human communication (design task, manual) - t-635: Auto-detect agent author via OMNI_AUTHOR env var - t-636: ...
Gerrit-like patch workflow for agent development
Build a Gerrit-inspired patch-based development workflow for coding agents. Core principle: one task = one commit. Revisions are amends (not new commits). Branches are ephemeral pointers to individual...
Agent debugger for Haskell agent
Build a debugger for the Haskell agent (Omni/Agent.*) only. Focus on prompt construction + hydration trace, tool calls, and inference events. No pi support for this epic. Goals: - Capture prompt buil...
Ava-Pi notification system
Enable ava to receive real-time notifications when pi agents complete tasks. Goal: replace polling-based heartbeat checks with push-based notifications via unix socket....
LLM Newsreader
Haskell RSS/Atom newsreader with LLM-powered topic clustering. Stack: SQLite (FTS5), req, servant, ollama (nomic-embed-text). Single-user, mobile-first, behind VPN. Design documented in code comments...
Free Monad Agent Architecture
Refactor the agent system to use a free monad over inference primitives, enabling parallel composition, pluggable state strategies, first-class observability, and checkpointing. See Omni/Agent/ARCHITE...
Agentd MVP - Minimal container runtime for agent workflows
## Summary Minimal viable container runtime for running agent workflows in isolated OCI containers. ## Goal Run markdown workflows in isolated containers in the omnirepo, for personal use. ## Non-...
Migrate Ava to run as ben user with dedicated worktree
Currently Ava runs as a separate 'ava' user, which causes many issues: - File permission problems - Git config/hooks don't work - Need symlinks in /usr/local/bin for tools - Can't use ben's pi token (...
Ava orchestrator improvements
Improvements to the Ava coding orchestrator (pi-orchestrate, pi-code, pi-review) identified during initial testing session....
Expand Ava's task management tools
Give Ava full access to task database functionality via LLM tools. Currently she only has work_on_task and list_ready_tasks. She needs tools for viewing, creating, updating, and querying tasks....
Design Ava Telegram interface for task orchestration
Design the Telegram conversation interface for Ava to orchestrate coding tasks. ## User Stories ### Start Work on Existing Task User: "work on t-123" Ava: "Starting t-123: Add --help flag..." [spawn...
Ava as orchestrator: idea → tasks → monitoring → notification
Goal: Drive the entire dev process from phone via Telegram. Discuss ideas with Ava, she creates tasks, monitors execution, notifies on issues. ## Current State - Ava exists as Telegram bot (Omni/Ava....
Automated process review and improvement
Goal: Regularly review completed work and suggest process improvements (AGENTS.md updates, new skills/tools). ## Current State - Work happens, gets committed - No systematic review of what worked/did...
Tighter pi + task integration
Goal: Make pi automatically integrate with the task system so agents claim tasks, update status, and log work. ## Current State - task tool exists (Omni/Task.hs) with create/update/show/ready command...
Improve dependency pinning system (deps/niv fork)
Improve the deps tool (niv fork) and the Nix dependency pinning infrastructure. Goals: 1. Add auto-override field to control which sources override nixpkgs packages 2. Fix version field inconsistency ...
Epic: Ava Tool Trace Viewer Mini-App
Add a Telegram mini-app that allows users to view full tool call traces from Ava's responses. When Ava uses tools, she can include a clickable link that opens an expandable view of the tool inputs/out...
Mini-PaaS Deployment System
Pull-based deployment system with S3 binary cache, manifest-driven systemd services, and Caddy reverse proxy. Enables deploying services without full NixOS rebuilds....
Ava Customer Acquisition Agent
Enable Ava (Telegram bot) to drive customer acquisition and launch support for PodcastItLater. ## Goal Validate or invalidate PodcastItLater hypothesis in 30 days via user conversations + revenue att...
Unified Task Timeline
Unify comments, activity, tool calls, and status changes into a single chronological timeline per task. ## Current Problem - Comments shown separately from agent activity - Agent Log is a separate pa...
Agent Observability & Event Streaming
Add structured event logging and streaming for agent work sessions, enabling real-time observation in the web UI and future interactive chat. ## Goals - Full visibility into agent reasoning and tool ...
Epic verification flow: require human review before closing epics
Epics need more verification before they are closed. Currently they auto-transition to Review when all children complete, but there's no mechanism to ensure the major features were actually shipped an...
Knowledge Base v1 - facts storage with web and CLI interface
Status Badge UX - merge badge and dropdown into single clickable element
Space Efficiency - denser layout with list-groups and inline labels
Epic Lifecycle - auto-review gate and epics overview page
Navbar v2 - horizontal layout with dropdowns and mobile collapse
Web UI: Task investigation features for humans
## Problem When a task needs human intervention, the web UI doesn't show enough context to understand what happened. Currently you can only see: - Task title/description - Current status - Retry cont...
Process: Ensure schema migrations run on deployment
## Problem Found t-148.1, t-148.2, t-148.3 were marked Done but the feature wasn't working because: 1. Code was written and deployed 2. But task_activity table didn't exist in production DB 3. jr tas...
Add HTMX for dynamic updates
## Problem Current web UI is fully static - every interaction requires full page reload. Want more dynamic feel without heavy JavaScript frameworks. ## Solution: HTMX HTMX allows HTML elements to m...
Design: Epics as design docs
## Problem Design docs are currently orphaned: - DB has design_docs table (empty, unused) - DB has design_doc column on tasks (not in Task record) - Task.taskDescription exists but isn't emphasized f...
Jr Web V2: Bug fixes and design refresh
Address functional bugs and redesign the UI for a denser, more Gerrit-like experience. ## Functional Bugs (P1) 1. Filter dropdowns fail on 'All' selection - returns empty string causing parse error ...
jr: Worker observability - store status updates in DB and show in web UI
Currently worker progress is only visible in CLI logs. Need to persist status updates to DB and display in web UI. Goals: - See what stage a worker is at (claiming, running amp, reviewing, retrying) ...
jr: Consolidate task command into jr
Remove standalone 'task' binary, use ~/.cache for db path, update all docs to reference 'jr task'...
Workout Agent: Training Planner
An agent that reads workout data from intervals.icu and creates training plans. Uses their API for both reads and writes....
hledger Agent: Personal Finance Assistant
An agent that reads hledger journals and answers questions about personal finances. Interface via email. Uses the Agent Engine....
Design Docs: Replace Epics with Planning Documents
Planning documents replace epics as the organizing unit. A design doc captures goal, solution, plan, and links to generated tasks. See _/llm/PLAN_Agent_Architecture.md...
Agent Engine: Replace Amp with Native Implementation
Extract a native LLM agent engine to replace amp subprocess dependency. See _/llm/PLAN_Agent_Engine_V2.md for full design. Key goals: 1. Engine.hs - Generic agent loop with pluggable tools 2. Tools....
jr: Planner Agent
Implement the planner agent role. The planner: 1) Reviews failed tasks (3 rejections) and replans them 2) Creates design docs and task breakdowns from goals 3) Has interactive chat mode for planning s...
jr: TOML Configuration File
jr: Cost Tracking
jr: Knowledge Base Foundation
MVP 1: Unified Jr Interface
Evolve the current task and agent tools into a unified 'jr' binary. This is MVP 1. Reference Plan: /home/ben/omni/_/llm/PLAN_Jr_Evolution.md...
PodcastItLater: Mailgun Integration
Phase 1: Foundations (Task & CI)
Cleanup Omni/Agent files and tests
Remove unused test files, migrate useful tests to the main suite, and remove legacy bash prototype scripts replaced by the Haskell implementation....
Codebase Health and Tooling Improvements
Collection of tasks to improve the robustness of the codebase (builds), the usability of the 'task' tool, and the accuracy of the agent's status reporting....
Improve Worker Autonomy
Multi-Agent System 2.0 (Haskell Agent)
PodcastItLater: UX Polish
Multi-Agent System 2.0
General Code Quality Refactor
Complete comprehensive test suite
PodcastItLater: Path to Paid Product
Task Manager Improvements