Telegram connector — read, process, and acknowledge Telegram bot messages as a task inbox for your AI agent
Slack connector — send notifications and read messages from configured channels
Unified command interface — all issue trackers share the same CLI commands: list, get, create, edit, start, delete, comment, status
95% fewer tokens than MCP servers — human extracts signal and discards noise, so the same Jira tickets that cost 5,781 tokens via raw API cost 267 through human
Credentials never touch the AI — API tokens are stored on your machine and used by the CLI; Claude Code never sees them
Quick commands — human get KAN-1 and human list --project=KAN auto-detect your tracker, no tracker prefix needed
Cross-tracker search — human search queries a local SQLite FTS5 index built from all connected trackers and Notion
Search index — human index builds and syncs a local full-text search index across all your tools
/human-ideate — challenge a rough idea with 5 forcing questions, then create a PM ticket in your tracker
/human-sprint — run the full pipeline in one command: ideate, plan, execute, review
/human-ready — evaluate a ticket against a Definition of Ready checklist before implementation starts
/human-brainstorm — explore implementation approaches and surface alternatives you might not have considered
/human-plan — fetch a ticket, explore the codebase, and produce a structured implementation plan
/human-execute — load a plan and execute it step by step with a review checkpoint at the end
/human-review — review the current branch’s changes against the ticket’s acceptance criteria
/human-done — evaluate whether the implementation is complete and shippable
/human-bug-plan — analyze a bug ticket for root cause and produce a fix plan
/human-findbugs — scan the codebase for logic errors, race conditions, and security issues with a multi-agent pipeline
/human-security — deep security audit with attack surface mapping, OWASP Top 10 scanning, and attack chain analysis
/human-gardening — analyze structural debt, duplication, and complexity hotspots with a health scorecard
Structured output in .human/ — all skill reports are markdown files saved to .human/ and tracked in git for auditability
Skill chaining — output from one skill feeds the next; /human-plan output is consumed by /human-execute
Works with any issue tracker — skills auto-fetch tickets from whichever tracker you’ve configured
Chrome Bridge — use claude --chrome inside devcontainers by tunneling Chrome DevTools Protocol to the host
MCP/OAuth in devcontainers — browser-based OAuth flows from inside containers just work, including Claude Code subscription linking
Firewall & HTTPS proxy — allowlist or blocklist domains for AI agent outbound network access using SNI-based filtering
.env secret protection — host-side FUSE filesystem filters .env files so AI agents see they exist but can never read the secrets inside
Open URLs from devcontainers — human-browser replaces $BROWSER inside containers and opens URLs on your host machine
Daemon mode — human daemon start runs a background daemon that forwards CLI commands, credentials, and browser access from containers to the host
Daemon token auth — token-based authentication ensures only authorized containers can communicate with the daemon
Live TUI dashboard — human tui shows token usage, running Claude Code instances, busy/ready state, and tracker issues in a live terminal UI
Per-model token breakdown — token usage is broken down by model family and version with per-instance bars
5-hour billing window tracking — automatically selects the current Claude Code billing window and shows what counts against your limit
Auto-discovery of Claude Code instances — finds every running instance on the host and inside Docker containers without configuration
Subagent visibility — busy instances show their running subagents with agent name, type, and elapsed time
Tracker pipeline state — the TUI shows open issues with pipeline state: Ready for Plan, Planning, Planned, Backlog, In Dev, Done
Policy-based operation control — block or require confirmation for specific operations like delete, assign, or transition:Done in .humanconfig.yaml
Safe mode — --safe flag blocks all destructive operations
Multiple output formats — JSON (default), markdown, and human-readable tables via --table
Multiple named instances per tracker — configure separate Jira instances for different teams or projects
Flexible credential resolution — tokens resolve from CLI flags, per-instance env vars, global env vars, or config file
Devcontainer feature — one-line install via ghcr.io/stephanschmidt/treehouse/human:1 in devcontainer.json
human init — interactive setup wizard for trackers, connectors, and daemon configuration
human install --agent claude — install all skills to .claude/skills/ in one command
Allowed Telegram users — restrict which Telegram users the bot listens to by user ID
Docker container auto-discovery — the TUI and daemon detect Claude Code running inside Docker containers via the Docker API
No MITM proxy — the HTTPS firewall reads SNI only, no certificate installation, no traffic decryption, data stays encrypted end-to-end
Defense in depth — credential isolation, .env filtering, and network firewalling combine for three layers of AI agent security