agentleFS
mercurialsolomercurialsolo/claudectl/docs/llms.txt

claudectl

llms.txt204 starsChanged 18 days agoMITView on GitHub

Hive mind for Claude Code - expand your consciousness with an agentic hive that learns from you once and remembers everywhere claudectl is a Rust TUI and daemon that supervises every Claude Code session on a machine, and a hive that joins machines together. A local LLM learns which tool calls you approve and deny; naming a hive makes that knowledge joinable, so a second machine starts with what the first one learned. It shows which agent is blocked, burning…

  • Reads credentials
  • Installs packages

We split this llms.txt into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read-only hive membership: a hive.read capability grant receives the hive knowledge and can never contribute any.

    line 39

≈ “Common” is an estimate: at least 1 in 5 of the 3,225 llms.txt files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
llms.txt
# claudectl

> Hive mind for Claude Code - expand your consciousness with an agentic hive that learns from you once and remembers everywhere

claudectl is a Rust TUI and daemon that supervises every Claude Code session on a machine, and a hive that joins machines together. A local LLM learns which tool calls you approve and deny; naming a hive makes that knowledge joinable, so a second machine starts with what the first one learned. It shows which agent is blocked, burning budget, or stalled, and lets you intervene without tab hunting.

## Install

```
brew install mercurialsolo/tap/claudectl
cargo install claudectl
```

Setup:
```
claudectl init      # onboarding wizard: budget, brain, hooks, bus role, skills
claudectl doctor    # verify install + runtime health
claudectl           # launch dashboard
claudectl demo      # try with fake sessions
```

## Key Features

- Live dashboard: status, burn rate, context %, CPU, memory, tokens, sparklines for all sessions
- One-key actions: approve (y), send input (i), launch new session (n), kill (d), record GIF (R)
- Budget enforcement: per-session and daily limits with auto-kill
- Local LLM brain: ollama/gemma auto-approves safe commands, denies dangerous ones, learns from corrections
- Auto-rules engine: TOML rules to approve/deny by tool name, command pattern, project, cost
- Health monitoring: detects stalls, context saturation, cache issues, cost spikes, retry loops
- Multi-session orchestration: dependency-ordered task graphs, prompt decomposition into parallel DAGs
- Event hooks: desktop notifications, shell commands, webhooks on session events
- Session recording: highlight reels as GIFs (edits, commands, errors - strips idle time)
- Relay & hive mind: connect instances across machines, share brain learnings peer-to-peer
- Remote task delegation: offload tasks to connected peers with status streaming
- Discovery: invite codes, word phrases, QR codes, LAN broadcast scanning
- Named hives: name a hive, advertise it over LAN, find it with hive discover, join it with a cctl://hive/ link, code or phrase
- Gossip: both ends of a relay connection sync incrementally every 12 seconds, so joins, reconnects and late arrivals all catch up; accepted units propagate onward up to a hop cap
- Host-enforced join policy: invite / ask / open; ask queues each request for the owner rather than admitting on possession of a link
- Read-only hive membership: a hive.read capability grant receives the hive knowledge and can never contribute any; refusals are sent on the wire
- Durable relay: relay install-agent runs it under launchd so it survives logout
- Terminal support: Ghostty, tmux, Kitty, Warp, iTerm2, Terminal.app, WezTerm, GNOME Terminal

## How It Works

Reads Claude Code's local files (no API keys, read-only). Nothing is sent anywhere until you turn it on: webhooks carry session metadata, the relay carries session snapshots and delegated tasks to paired peers, the hive carries knowledge units to hive members, and the optional Jev classifier on the query surface carries a third party's question plus one paragraph of CLAUDE.md (needs TYPESAFE_API_KEY; [query] jev_enabled = false is the hard off). Sources:
- ~/.claude/sessions/*.json - session metadata
- ~/.claude/projects/{slug}/*.jsonl - conversation logs with token usage
- ps - CPU%, memory, TTY per process

Status inference uses multiple signals: waiting_for_task events, CPU thresholds, stop_reason fields, message recency.

## Architecture

~6.3 MB binary with default features (bus+coord+relay+hive), ~3.5 MB sync-only. Sub-50ms startup, 7 runtime crates in the sync core, no async runtime except the two MCP stdio servers (claudectl bus stdio, claudectl query stdio), which each build their own current-thread Tokio runtime.

Cargo workspace, dependencies flow strictly downward: claudectl (binary) -> claudectl-tui -> claudectl-core. The UI/runtime contract is a set of traits in crates/claudectl-core/src/runtime.rs, implemented over the real subsystems by src/runtime/.

claudectl-core (crates/claudectl-core/src): session.rs, discovery.rs, monitor.rs, process.rs, transcript.rs, history.rs, fleet.rs, forecast.rs, health.rs, rules.rs, hooks.rs, launch.rs, models.rs, skills.rs, theme.rs, logger.rs, helpers.rs, config.rs, runtime.rs, terminals/
claudectl-tui (crates/claudectl-tui/src): app/, ui/ (table, detail, help, status_bar, peers, skills, supervisor, demo_tour), recorder.rs, session_recorder.rs, demo.rs
Binary (src): main.rs, commands.rs, config.rs, orchestrator.rs, doctor.rs, ingest.rs, brain_screen.rs, rate_limit.rs, team_policy.rs, init/, runtime/
Brain: engine.rs, client.rs, context.rs, decisions.rs, retrieval.rs, preferences.rs, pref_store.rs, outcomes.rs, sequences.rs, detectors.rs, heuristic.rs, risk.rs, review.rs, audit.rs, autopsy.rs, baseline.rs, briefing.rs, garden.rs, health.rs, agents.rs, mailbox.rs, prompts.rs, evals.rs, insights.rs, metrics/
Coord / supervisor: supervisor.rs (pure reconciler), actuator.rs, verify.rs, resume.rs, tasks.rs, store.rs, events.rs, exporter.rs, session_policy.rs, hook_events.rs, promotion.rs, injection.rs, interrupt_bus.rs, metrics.rs, pr.rs, evals.rs, adapter*.rs, supervisor_cli.rs, cli.rs
Bus: store.rs, roles.rs, policy.rs, mcp.rs, cli.rs
Relay: mod.rs, cli.rs, crypto.rs, protocol.rs, peer.rs, listener.rs, mesh.rs, hivesync.rs, hivejoin.rs, delegation.rs, tasks.rs, outcome.rs, worker.rs, transfer.rs, transient.rs, advertise.rs, invite.rs, lan.rs, http.rs, agent.rs
Hive: mod.rs, cli/, store.rs, archive.rs, distiller.rs, merger.rs, gossip.rs, trust.rs, injection.rs, identity.rs, membership.rs, accept.rs, discovery.rs, exposure.rs, convergence.rs, effectiveness.rs, feedback.rs
Access: mod.rs, cli.rs, scope.rs, token.rs, grant.rs
Context index: mod.rs, git.rs, deny.rs, docs.rs, module_map.rs, exposure.rs
Query: mod.rs, core.rs, rank.rs, http.rs, mcp.rs, cli.rs, jev.rs, classify.rs, thresholds.rs, escalate.rs, spend.rs
Terminals: ghostty, kitty, tmux, wezterm, warp, iterm2, terminal_app, gnome_terminal, windows_terminal

## Docs

- [Quick Start](https://mercurialsolo.github.io/claudectl/quickstart/)
- [Configuration](https://mercurialsolo.github.io/claudectl/configuration/)
- [Reference](https://mercurialsolo.github.io/claudectl/reference/)
- [Terminal Support](https://mercurialsolo.github.io/claudectl/terminal-support/)
- [Troubleshooting](https://mercurialsolo.github.io/claudectl/troubleshooting/)
- [Fleet Supervisor](https://mercurialsolo.github.io/claudectl/fleet-supervisor/)
- [Relay & Hive Mind](https://mercurialsolo.github.io/claudectl/relay/)
- [Capability Grants](https://mercurialsolo.github.io/claudectl/access/)
- [Open Cluster](https://mercurialsolo.github.io/claudectl/open-cluster/)
- [Contributing](https://mercurialsolo.github.io/claudectl/contributing/)

## Design specs

- [Agent Bus](https://mercurialsolo.github.io/claudectl/AGENT_BUS/)
- [Relay & Hive](https://mercurialsolo.github.io/claudectl/relay-and-hive/)
- [Relay Discovery](https://mercurialsolo.github.io/claudectl/relay-discovery/)
- [Hive Storage](https://mercurialsolo.github.io/claudectl/hive-storage/)

## Links

- Source: https://github.com/mercurialsolo/claudectl
- License: MIT

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.