skills 🧠

The agent skill collection of Kevin Veen-Birkenbach. Every skill is pinned in skills-lock.json for reproducible installs. First-party skills live under skills/ and are layered on top of the locked ones at install time. Compatible with Claude Code, Codex, Gemini CLI, Cursor, Copilot, Windsurf, Cline, and other agents.

Naming 🏷️

Every first-party skill is named kvb-<category>-<name> with one of these categories: think, create, code, ci, git, release, check, org, agent, skill, coach: think thinks without producing a file, create produces an artifact, code works on code, ci on pipelines, git on history and forges, release announces a release (changelog, article, posts), check analyses and reports findings, org coordinates agents, agent acts on one agent itself: its repository setup (code index, agent instructions) and its state (drain, suspend, resume), skill works on skills themselves, coach serves the operator's wellbeing. Only kvb-cleanup, kvb-coach, kvb-help and the collectors kvb-dream, kvb-optimize, kvb-release and kvb-transcend stand without a category. Names use a-z, 0-9 and single hyphens, at most 64 characters, and equal their directory name. tests/test_naming.py enforces all of it, including that the organigram below lists every skill under its category.

Organigram 🏛️

Every first-party skill under its category.

flowchart LR
  kvb((kvb))
  kvb --> think[think]
  kvb --> create[create]
  kvb --> code[code]
  kvb --> ci[ci]
  kvb --> git[git]
  kvb --> release[release]
  kvb --> check[check]
  kvb --> org[org]
  kvb --> agent[agent]
  kvb --> skill[skill]
  kvb --> coach[coach]
  kvb --> none[no category]
  think --> think_listen[kvb-think-listen]
  think --> think_explain[kvb-think-explain]
  think --> think_dialectic[kvb-think-dialectic]
  think --> think_criticize[kvb-think-criticize]
  think --> think_brainstorm[kvb-think-brainstorm]
  think --> think_gaps[kvb-think-gaps]
  create --> create_plan[kvb-create-plan]
  create --> create_requirement[kvb-create-requirement]
  code --> code_scope_guard[kvb-code-scope-guard]
  code --> code_clean[kvb-code-clean]
  code --> code_refactor[kvb-code-refactor]
  code --> code_test[kvb-code-test]
  code --> code_tune[kvb-code-tune]
  code --> code_transcend[kvb-code-transcend]
  ci --> ci_triage[kvb-ci-triage]
  ci --> ci_artifacts[kvb-ci-artifacts]
  ci --> ci_observe[kvb-ci-observe]
  git --> git_commit_next[kvb-git-commit-next]
  git --> git_commit_no_verify[kvb-git-commit-no-verify]
  git --> git_merge_fix[kvb-git-merge-fix]
  git --> git_issue_close[kvb-git-issue-close]
  git --> git_issue_solve[kvb-git-issue-solve]
  git --> git_pr_create[kvb-git-pr-create]
  git --> git_pr_label[kvb-git-pr-label]
  git --> git_update_branches[kvb-git-update-branches]
  git --> git_branch_gc[kvb-git-branch-gc]
  git --> git_orphans_find[kvb-git-orphans-find]
  release --> release_changelog[kvb-release-changelog]
  release --> release_article[kvb-release-article]
  release --> release_post[kvb-release-post]
  git --> git_hotfix[kvb-git-hotfix]
  git --> git_push[kvb-git-push]
  git --> git_diff[kvb-git-diff]
  check --> check_legal[kvb-check-legal]
  check --> check_security[kvb-check-security]
  check --> check_principles[kvb-check-principles]
  check --> check_confidence[kvb-check-confidence]
  check --> check_break_risk[kvb-check-break-risk]
  check --> check_gate[kvb-check-gate]
  check --> check_efficiency[kvb-check-efficiency]
  check --> check_speed[kvb-check-speed]
  check --> check_ansible[kvb-check-ansible]
  check --> check_score[kvb-check-score]
  org --> org_organize[kvb-org-organize]
  org --> org_swarm[kvb-org-swarm]
  org --> org_unify[kvb-org-unify]
  org --> org_coordinate[kvb-org-coordinate]
  org --> org_perform[kvb-org-perform]
  org --> org_matrix[kvb-org-matrix]
  org --> org_rasci[kvb-org-rasci]
  org --> org_artefacts[kvb-org-artefacts]
  org --> org_wsjf[kvb-org-wsjf]
  org --> org_robot[kvb-org-robot]
  org --> org_operator[kvb-org-operator]
  org --> org_priority[kvb-org-priority]
  org --> org_drain[kvb-org-drain]
  org --> org_suspend[kvb-org-suspend]
  org --> org_resume[kvb-org-resume]
  think --> think_choice[kvb-think-choice]
  org --> org_radio[kvb-org-radio]
  org --> org_queue[kvb-org-queue]
  org --> org_queue_estimate[kvb-org-queue-estimate]
  org --> org_update[kvb-org-update]
  agent --> agent_optimize[kvb-agent-optimize]
  agent --> agent_drain[kvb-agent-drain]
  agent --> agent_suspend[kvb-agent-suspend]
  agent --> agent_resume[kvb-agent-resume]
  skill --> skill_tune[kvb-skill-tune]
  skill --> skill_auto[kvb-skill-auto]
  skill --> skill_essence[kvb-skill-essence]
  skill --> skill_transcend[kvb-skill-transcend]
  skill --> skill_compile[kvb-skill-compile]
  skill --> skill_cleanup[kvb-skill-cleanup]
  skill --> skill_silence[kvb-skill-silence]
  none --> cleanup[kvb-cleanup]
  none --> help[kvb-help]
  none --> transcend_all[kvb-transcend]
  none --> dream_all[kvb-dream]
  none --> optimize_all[kvb-optimize]
  none --> release_all[kvb-release]
  none --> coach_all[kvb-coach]
  coach --> coach_meditation[kvb-coach-meditation]

Skill map 🗺️

How the first-party skills work together. A solid arrow means the skill invokes or relies on the other one; a dotted arrow means it fires automatically. Every skill that reports a result closes with the kvb-check-confidence block; those arrows are left out. The locked third-party skills (caveman, ponytail, archify) are not drawn.

flowchart LR
  subgraph triggers [Triggers]
    commit_req([commit request])
    stage_req([staging or fix ready])
    edit_hook([edit in a git worktree])
    reminders([12 h agent hint, 17:00 GNOME popup])
    full_hour([full hour])
    code_task([code task])
    spawn([subagent spawn])
  end
  subgraph cat_think [think]
    active_listening[kvb-think-listen]
    explain[kvb-think-explain]
    dialectic[kvb-think-dialectic]
    criticize[kvb-think-criticize]
    brainstorm[kvb-think-brainstorm]
    gaps[kvb-think-gaps]
    choice[kvb-think-choice]
  end
  subgraph cat_create [create]
    plan[kvb-create-plan]
    requirement[kvb-create-requirement]
  end
  subgraph cat_code [code]
    scope_guard[kvb-code-scope-guard]
    clean[kvb-code-clean]
    refactor[kvb-code-refactor]
    test[kvb-code-test]
    tune[kvb-code-tune]
    code_transcend[kvb-code-transcend]
  end
  subgraph cat_ci [ci]
    triage[kvb-ci-triage]
    ci_artifacts[kvb-ci-artifacts]
    observe[kvb-ci-observe]
  end
  subgraph cat_git [git]
    commit_next[kvb-git-commit-next]
    commit_no_verify[kvb-git-commit-no-verify]
    merge_fix[kvb-git-merge-fix]
    issue_close[kvb-git-issue-close]
    issue_solve[kvb-git-issue-solve]
    pr_create[kvb-git-pr-create]
    pr_label[kvb-git-pr-label]
    update_branches[kvb-git-update-branches]
    branch_gc[kvb-git-branch-gc]
    orphans_find[kvb-git-orphans-find]
    hotfix[kvb-git-hotfix]
    git_push[kvb-git-push]
    git_diff[kvb-git-diff]
  end
  subgraph cat_release [release]
    changelog[kvb-release-changelog]
    release_article[kvb-release-article]
    release_post[kvb-release-post]
  end
  subgraph cat_check [check]
    advocate[kvb-check-legal]
    medusa[kvb-check-security]
    sceptic[kvb-check-principles]
    confidence[kvb-check-confidence]
    break_risk[kvb-check-break-risk]
    gate[kvb-check-gate]
    efficiency[kvb-check-efficiency]
    speed[kvb-check-speed]
    check_ansible[kvb-check-ansible]
    score[kvb-check-score]
  end
  subgraph cat_org [org]
    democracy[kvb-org-organize]
    funnel[kvb-org-swarm]
    unify[kvb-org-unify]
    coordinate[kvb-org-coordinate]
    subagent_model[kvb-org-perform]
    status_matrix[kvb-org-matrix]
    rasci[kvb-org-rasci]
    artefacts[kvb-org-artefacts]
    wsjf[kvb-org-wsjf]
    robot[kvb-org-robot]
    operator[kvb-org-operator]
    priority[kvb-org-priority]
    org_drain[kvb-org-drain]
    org_suspend[kvb-org-suspend]
    org_resume[kvb-org-resume]
    radio[kvb-org-radio]
    queue[kvb-org-queue]
    queue_estimate[kvb-org-queue-estimate]
    org_update[kvb-org-update]
  end
  subgraph cat_agent [agent]
    agent_optimize[kvb-agent-optimize]
    agent_drain[kvb-agent-drain]
    agent_suspend[kvb-agent-suspend]
    agent_resume[kvb-agent-resume]
  end
  subgraph cat_skill [skill]
    autotune[kvb-skill-tune]
    autoskill[kvb-skill-auto]
    essence[kvb-skill-essence]
    skill_transcend[kvb-skill-transcend]
    compile[kvb-skill-compile]
    skill_cleanup[kvb-skill-cleanup]
    silence[kvb-skill-silence]
  end
  subgraph cat_none [no category]
    cleanup[kvb-cleanup]
    help_skills[kvb-help]
    transcend[kvb-transcend]
    dream[kvb-dream]
    optimize[kvb-optimize]
    release_all[kvb-release]
    coach[kvb-coach]
  end
  subgraph cat_coach [coach]
    meditation[kvb-coach-meditation]
  end

  commit_req -.-> commit_next
  commit_req -.-> commit_no_verify
  stage_req -.-> confidence
  edit_hook -. council gate .-> democracy
  reminders -.-> dream
  full_hour -.-> coach
  code_task -.-> clean
  code_task -.-> scope_guard
  spawn -.-> subagent_model

  commit_next --> merge_fix
  commit_next --> commit_no_verify
  commit_next --> explain
  commit_next --> dialectic
  issue_close --> confidence
  issue_solve --> issue_close
  issue_solve --> funnel
  issue_solve --> pr_create
  pr_create --> issue_close
  pr_create --> pr_label
  pr_create -. push .-> operator
  update_branches --> active_listening
  update_branches -. conflicts .-> merge_fix
  update_branches -. pushes .-> operator
  commit_no_verify --> clean
  commit_next --> clean
  clean --> sceptic
  clean --> refactor
  commit_no_verify --> advocate
  commit_next --> git_diff
  commit_no_verify --> git_diff
  advocate --> dialectic
  confidence --> break_risk
  confidence --> gate
  pr_label --> gate
  stage_req -.-> score
  score --> confidence
  score --> break_risk
  score --> efficiency
  score --> speed
  speed --> efficiency
  efficiency -. heavy commands .-> queue
  gate --> democracy
  gate --> hotfix
  triage --> hotfix
  democracy --> hotfix
  hotfix --> rasci
  confidence -. objective .-> autotune

  coordinate --> democracy
  cleanup --> democracy
  cleanup --> orphans_find
  branch_gc --> democracy
  orphans_find --> branch_gc
  orphans_find --> choice
  orphans_find --> operator
  funnel --> democracy
  funnel --> dialectic
  funnel --> subagent_model
  dialectic --> subagent_model
  democracy --> merge_fix
  democracy -. claim and release .-> rasci
  funnel --> rasci
  unify --> funnel
  unify --> rasci
  unify --> gate
  unify --> branch_gc
  unify --> cleanup
  unify --> radio
  status_matrix --> rasci
  cleanup --> artefacts
  artefacts --> rasci
  democracy --> wsjf
  wsjf --> rasci
  priority --> rasci
  priority -. pins outrank .-> wsjf
  operator --> rasci
  dream --> rasci
  org_update -. newsletter .-> radio

  observe --> triage
  observe --> ci_artifacts
  triage --> ci_artifacts
  triage --> democracy
  triage --> dialectic
  triage --> funnel
  triage --> tune
  tune --> democracy
  tune --> funnel
  tune --> merge_fix
  tune --> dialectic
  tune --> check_ansible
  check_ansible -. heavy commands .-> queue

  dialectic --> active_listening
  criticize --> dialectic
  criticize --> active_listening
  sceptic --> dialectic
  gaps --> sceptic
  gaps --> active_listening
  gaps --> requirement
  requirement --> active_listening
  requirement --> brainstorm
  requirement -. offer .-> plan
  brainstorm --> dialectic

  plan --> robot
  plan --> active_listening
  robot --> dialectic

  dream --> dialectic
  dream --> optimize
  dream --> cleanup
  dream --> orphans_find
  optimize --> skill_cleanup
  optimize --> transcend
  optimize --> autoskill
  optimize --> autotune
  optimize --> compile
  optimize --> essence
  optimize --> refactor
  optimize --> tune
  optimize --> silence
  silence -. peer protocol .-> democracy
  optimize --> agent_optimize
  optimize --> gaps
  agent_optimize -. install .-> operator
  agent_optimize -. domain graphs .-> rasci
  org_drain -. message .-> agent_drain
  org_suspend -. message .-> agent_suspend
  org_resume -. message .-> agent_resume
  agent_drain --> agent_suspend
  org_suspend -. hold and accept .-> operator
  org_resume -. hold and accept .-> operator
  agent_suspend -. stops .-> robot
  coach --> active_listening
  coach --> operator
  coach --> rasci
  coach --> meditation
  meditation --> active_listening
  autoskill --> autotune
  autotune --> essence
  transcend --> skill_transcend
  transcend --> code_transcend
  release_all --> changelog
  release_all --> release_article
  release_all --> release_post
  release_all --> active_listening
  release_all --> brainstorm
  release_article --> active_listening
  release_article --> brainstorm
  release_post --> active_listening
  release_post --> brainstorm
  changelog --> active_listening
  changelog --> brainstorm
  skill_transcend --> active_listening
  skill_transcend --> autotune
  code_transcend --> active_listening
  autotune -. own scripts .-> skill_transcend
  autotune -. deterministic prose .-> compile
  autoskill -. deterministic steps .-> compile
  skill_transcend -. deterministic moves .-> compile
  essence -. procedures .-> compile
  compile --> active_listening
  skill_cleanup --> operator
  skill_cleanup -. unknown skills .-> transcend
  tune -. measurements .-> code_transcend

  test --> triage
  test --> dialectic
  test --> clean
  refactor --> clean

Requirements 📦

• Node.js (provides npx skills).
• jq (only required for make update).
• Python jinja2 for make test (pinned in orchestrator/requirements.txt).
• Docker with the compose plugin (optional, runs the Agent Orchestrator).

Installation 🚀

Install the skills globally into your home (~/.agents/skills and ~/.claude/skills):

make install

Copy the skills into a specific project (<repo>/.agents/skills and <repo>/.claude/skills):

make project TARGET=/path/to/repo

Both commands also patch the target's .claude/settings.local.json — the machine-local, untracked file, so a project's reviewed .claude/settings.json is never rewritten by an install: the caveman and ponytail plugins are enabled so their modes auto-activate on session start, and two reminder hooks are registered — kvb-dream points each agent at /kvb-dream at most once every twelve hours (stamped per session in ~/.claude/.dream-stamps/), and kvb-coach runs its short form at the first prompt of every full hour, once across all agents (stamped in ~/.claude/.coach-hours/); the global install also enables the kvb-dream-notify.timer systemd user timer, which recommends /kvb-dream in a GNOME notification daily at 17:00 local time and catches up once after a missed run. A PreToolUse gate from the kvb-org-organize skill blocks edits to files in a git worktree until the session holds a claim (skills/kvb-org-organize/scripts/council.py), and its resource watch (skills/kvb-org-organize/scripts/watch.py) measures disk, Docker and caches, cleans up low-risk data and blocks new heavy jobs before the disk runs out; it stays inactive until ~/.claude/agents-council/config.json exists, and because its hooks run that config's commands outside the sandbox, the install adds Edit(~/.claude/agents-council/**) to permissions.deny, which also covers Write in every permission mode. The kvb-org-radio skill registers five hooks: its prompt hook and a PostToolBatch hook deliver the immediate radio (call, broadcast, security, pan-pan, mayday) mid-turn, a Stop hook delivers the post (post, newsletter and call receipts) when the turn ends, an asyncRewake Stop hook wakes an idle session for the radio, and a PreToolUse hook refuses every writing action while a mayday holds, except those of its sender and of the agents that join it with radio.py assist. The PreToolUse hook of the kvb-org-queue skill sends heavy commands through queue.py hold once the config sets queue_enforce, and holds a held command to the settings' Bash deny and ask rules, which see only queue.py hold. The global install also installs the MEDUSA security scanner with pipx for kvb-check-security, unless medusa is already on the PATH. Two hooks of the kvb-org-perform skill (skills/kvb-org-perform/scripts/log.py) log every Agent spawn with its task class, model and effort, and store the quality feedback a sandboxed session prints, in ~/.local/state/agents-council/subagent-model.jsonl. Every hook runs /usr/bin/python3 -I, so neither the python3 of a sandbox-writable virtualenv on the PATH nor PYTHONPATH reaches the hooks, which run outside the sandbox; older bare python3 hook entries are rewritten. The reminder hooks run /usr/bin/bash and pin PATH=/usr/bin:/bin, and the council scripts keep only the PATH directories this uid cannot write (skills/kvb-org-organize/scripts/hookpath.py), so no binary an agent plants on the inherited PATH runs outside the sandbox. Existing settings are preserved.

The global install also runs the Agent Orchestrator, a local review page in which the operator approves or rejects the agents' staged commits (the kvb-git-diff skill): it copies orchestrator/ to ~/.local/share/kvb-orchestrator, starts it there with docker compose up -d --build (restart unless-stopped, bound to 127.0.0.1 only, state in ~/.local/state/kvb-orchestrator) and registers its MCP server for Claude Code at user scope. Its port is KVB_ORCHESTRATOR_PORT in orchestrator/.env; the hub page is http://localhost:<port>/, the MCP endpoint /mcp. Without docker the hub is skipped and the commit skills ask in chat. The hub reads new radio traffic aloud in maritime phraseology and takes push-to-talk input; its image builds both engines from pinned, sha256-verified sources and runs each as a separate process: Piper (piper-tts, GPL-3.0-or-later) with the en_US-libritts-high voice (904 speakers, trained on LibriTTS, CC BY 4.0, from rhasspy/piper-voices) and whisper.cpp (MIT) with the multilingual ggml-base-q8_0 model (OpenAI Whisper weights, MIT, 82 MB). The operator's own voice is KVB_ORCHESTRATOR_OPERATOR_VOICE in orchestrator/.env. Its Tree reads only the file pkgmgr branch tree writes: the install clones PKGMGR_REPO (default the GitHub repository, never the sandbox-writable local checkout), builds the newest v* tag a fully trusted GPG key signed into ~/.local/share/kvb-orchestrator/pkgmgr, and enables the kvb-orchestrator-tree.timer user timer, which writes ~/.local/state/pkgmgr/tree/branch-tree.json a minute after its previous run while the hub page polls, on at most half a core; the hub mounts that directory read-only and starts no git. A signed tag without branch tree leaves the timer off.

Stream → Radio (#stream/radio) is the hub's radio interface. Its station list shows one line per station, the full name in the tooltip, with its state and a play or stop button; when the browser blocks autoplay, the stream fails or the CSP refuses its origin, the row and a toast say so, and one click starts the station. New stations play daily around the clock. While the hub reads a message aloud, the speaking agent pulses there and in the Orbit and Chord views, under the caption " speaking" with the message's first line; under prefers-reduced-motion the ring stays still. The composer sends on all seven channels: post and call to one agent, newsletter, broadcast, pan-pan, security and mayday to all. "Record" records with MediaRecorder; the hub transcribes the recording with whisper.cpp off the request thread (at most 120 s of audio, a 120 s timeout, four at a time), keeps it for replay in ~/.local/state/kvb-orchestrator/voice (the newest 50) and puts the transcript into the field to edit; the message goes out as text, with the recording's host path and sha256 only when "Attach recording" is ticked. The hub transcribes at most six recordings a minute and 600 s of audio an hour. Every operator message is passkey-signed, and the hub verifies the signature against the pinned keys, each once, before it relays; operator_message(id, text, to) answers match only for the signed text and recipient; for a denial the signature covers the recipient the inbox resolved at signing, and the answer's text is only the signed reason, never the agent's own words the radio row quotes. "Cancel mayday", Evening, Sunset, Sunrise and a settings save that adds a stream URL take the passkey the same way.

The hub page loads nothing from the network: its icons come from Bootstrap Icons 1.13.1 (orchestrator/vendor/bootstrap-icons, MIT), inlined per page as the symbols it uses, and its 3D views from 3d-force-graph 1.80.1 and three 0.186.1 (orchestrator/vendor, MIT, served by the hub under /vendor/). In the Artifacts Tree, "Delete" on a merged-clean branch without skip reason, or "Delete all merged", shows the dry-run lines and asks for the passkey signature; the hub then writes a job (repository, branch, worktree and tip per item, no command) into ~/.local/state/kvb-orchestrator/cleanup/jobs. The kvb-orchestrator-cleanup.path user unit runs client/cleanup.py on the host, which verifies the signature against the pinned keys, runs pkgmgr branch cleanup --json --local --branch <name> as a dry run in the repository, then --apply only when that plan names exactly the signed branch as merged-clean at the signed tip and worktree, refuses everything else unchanged, writes each result to cleanup/results and rewrites the tree file. The buttons stay disabled while the tree file says delete_ready: false.

The Agents view "Hierarchy" draws the council's formal offices as a collapsible org chart: Operator, chair, dimensions with their line representatives, organisers (cluster/<slug>) and presidents (work/<slug>); the chair in every graph is the holder of the chair office, not a fixed name. Appointing, handing over or recalling an office there asks for the passkey signature; the hub writes a job (office, signed holder, successor session, work; no command) into ~/.local/state/kvb-orchestrator/offices/jobs, and the kvb-orchestrator-offices.path user unit runs client/offices.py on the host, which verifies the signature, refuses a council directory others can write, a link on the path or a holder that changed since the signature. For a takeover or recall it moves the office aside and removes it only while the claim there still names the signed holder; an appoint or takeover then runs a plain council.py claim <office> for the successor by argv. council.py itself has no flag that overrides a live holder. The same view holds the operator's agent order as a drag list with up and down buttons: rasci.py agents order stores it in agent-order.json beside the register, where it breaks ties of equal task priority and orders the chair's grants without touching any task's --prio pin.

The Tasks tab (key 4, #tasks/list, #tasks/board, #tasks/rasci) shows the kvb-org-rasci register as a list, a board by status and a RASCI matrix; queue jobs from council/queue.json appear as "waiting for a grant", and the filter "Waits on Operator" is the inbox, which the Inbox entry under Stream opens: every pending review, ask, rule request, operator row and choice card, with no Choices column of its own. The Tasks toolbar is one row that wraps on narrow screens (views, filter, estimation, sort and business value, trend on the burn view), the status chips the second; every list row shows its ETA, and #tasks/3d draws the tasks and their dependencies on the 3D engine. Its "Request choices" dropdown has one button per channel, post and call to the agents ticked in it, newsletter, broadcast, pan-pan, Sécurité and mayday to all, that sends the fixed request "send me your open decisions as choices" without a text field, one passkey assertion over every recipient (/api/choices/sign, then /api/choices). Agents answer with decision_request, shown as choice cards, and a click on an option is the passkey-signed answer: after the 3 s undo the hub verifies the signature and posts the answer to the asking agent, whose decision_wait returns it as message and answer for operator_message. A task's sheet pins its priority (rasci.py set <task> --prio <n>) or sets and drops the operator's WSJF estimate (rasci.py estimate <task> Operator --value <fib> --complexity <fib> or --drop) after a passkey signature; the hub writes an argv-only job, and the runner, started with --rasci <rasci.py> --council <council root>, verifies it like an operator command and runs rasci.py from / without a shell, killing it after 30 s. Agent order stays untouched. The install copies the fresh runner unit and restarts the runner when it is enabled, and only while the installed council.py, nofollow.py and rasci.py match this tree: otherwise it disables the runner and asks for make install. The runner unit pins PATH to /usr/bin:/bin, so a signed command names any other program by its full path; a runner started without --rasci and --council keeps running signed commands and refuses task changes, and so does one whose council is a symlink or another uid's. Each worktree in the tree file may carry "flow": {"pull": [...], "push": [...], "source": "file"} or null, which the Tree, the Chord edges, the detail panel and the matching task show; a malformed flow shows as invalid. Worktrees and tasks with open work (uncommitted, unpushed, unmerged, task not done) carry a warning bar.

Without reinstalling the skills, make orchestrator-install installs, builds and starts only the hub, make orchestrator-check checks the installed hub (image, container, logs, HTTP, pinned keys, hook, runner, tree timer, disk; exit 1 on a failure), make orchestrator-pin CREDENTIAL=<blob> pins the passkey that "Register passkey" shows, make orchestrator-runner enables the runner service, and make orchestrator-hook registers the PermissionRequest hook in ~/.claude/settings.json, replacing an earlier Agent Orchestrator entry and keeping the file mode, after a copy to settings.json.before-kvb-orchestrator-hook (Claude Code reads ~/.claude/settings.local.json only in sessions started in the home directory). Agents that reach the Docker socket are root-equivalent: until the agent VMs of core requirement 093 isolate them, the signatures do not hold against such an agent. make watch-check shows whether the resource watch hook is stuck (running hooks, holders of the kept queue history, age of the stored stage).

For approvals the hub is only a relay: every approval of an ask, a staged commit or an operator command is signed end to end with a WebAuthn passkey held in KeePassXC (through KeePassXC-Browser, relying party localhost, origin http://localhost:<port>; ES256, EdDSA and RS256, so a hardware key can be pinned beside it). To register, open the hub at http://localhost:<port>/ (WebAuthn needs the domain; on 127.0.0.1 the page links there), choose "Register passkey" in the settings and run the host command the page shows, python3 ~/.local/share/kvb-orchestrator/client/approval.py pin --port <port> --approvers ~/.config/kvb-orchestrator/approvers.json --credential '<blob>', in a terminal outside every sandbox: it pins the key in ~/.config/kvb-orchestrator/approvers.json with mode 0600, a file no sandboxed agent can write and the hub mounts read-only; the hub never pins. Each approval, or one bulk selection, is one signature over the hashes of every item, a nonce and an expiry; each auto-approval rule is signed once. In the inbox, an operator command run with "Run" shows the runner's output in a console under its row until "Close". "Deny" on an ask, an operator row or a rule request signs the denial with its channel and optional reason (a hash no approval shares) and, after the 3 s undo, tells the requesting agent by post or by the radio class picked, never mayday; the hub verifies that signature against the pinned keys before it stores or sends anything, the hook passes the reason on only when it verifies the signature too, and a denied operator row ends as stopped. Agents ask for an auto-approval rule with the MCP tool rule_request (the rules form's limits; a request stays open for the ask lifetime of one hour, at most 20 in all, and 3 per agent name, which only spreads the queue since MCP names no caller); its row shows a dry run over the last 7 days, and only "Accept" with the passkey stores it. The hub stores no rule, of any kind, whose passkey assertion it cannot verify. The executors verify it themselves against the pinned keys and fail closed: the runner (client/runner.py, the kvb-orchestrator-runner user service) runs only signed commands, each signature once, recorded in ~/.local/state/kvb-orchestrator-approvals/runner.json; the PermissionRequest hook (client/permission_hook.py --port <port> --timeout <seconds> --approvers ~/.config/kvb-orchestrator/approvers.json --replay ~/.local/state/kvb-orchestrator-approvals/hook.json) allows only a signed item that matches its own tool, command, cwd and session, else it leaves the decision to Claude Code's own prompt; the commit skills run approval.py verify on the staged tree before committing, else they ask in the terminal. With no key pinned, every executor refuses; review rejections and a decision a rule takes stay unsigned. A signed rule never covers a command holding ;, &, |, a backtick, $, <, >, parentheses or a line break: such a command always asks. The hub page is trusted while it asks for a signature or shows the blob to pin: whoever controls the hub container can show one item and have another signed, or hand over a key of its own. Any local process, sandboxed agents included, can also send the unsigned operator actions (the bare decision choice /api/decisions/<id>/choose, which sends no answer, the priority order, settings that add no stream origin, review rejections, the bare ask denial /api/asks/<id>/deny, console dismissals) with the hub's guard headers, and the container reads the host /tmp for the review pages in /tmp/kvb-git-diff. All three close with the agent VMs of core requirement 093, like the Docker-socket gap.

Restart your agent afterwards so it loads the new skills.

Maintenance 🔧

Refresh every skill to its latest version and rewrite skills-lock.json:

make update

Validate skills-lock.json and shellcheck the scripts (also run by CI on every push and after every update):

make test

Related Tools 🧰

This collection is part of a wider toolset for managing IT and infrastructure:

• alias — shell aliases for common commands.
• package-manager — pkgmgr, a single command to install and manage these repositories.
• infinito-nexus — automation platform for self-hosted infrastructure.

License 📜

MIT License, Copyright (c) 2026 Kevin Veen-Birkenbach.

Author 👤

Kevin Veen-Birkenbach kevin@veen.world

Thanks 🙏

Thanks to the #ai:infinito.nexus community, and especially to Paulinhx for the inspiration.

Donation ❤️

If these tools help you, consider supporting their development:

• GitHub Sponsors
• Patreon
• Buy Me a Coffee
• PayPal
• Bitcoin: bc1qle7j8m9kvny2h2q3jzsnc64s4f5zme5re23pzc2lc2d7hhasc4dqachflp