Install
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/dotdevdotdev/agentwire-dev/agentwire-clinpx skills add dotdevdotdev/agentwire-dev --skill agentwire-cligit clone --depth 1 https://github.com/dotdevdotdev/agentwire-dev Per session 88 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 4,947 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00088 | $0.04947 |
| Opus 5 | $0.00044 | $0.02474 |
| Sonnet 5 | $0.00018 | $0.00989 |
| Haiku 4.5 | $0.00009 | $0.00495 |
Security
Grade C, and why
agentwire-cli scanned grade C with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
# DESTRUCTIVE: rm -rf's the worktree dir + new branch .claude/skills/agentwire-cli/SKILL.md · 343 lines
How it starts
The opening of the file, as written. The whole thing — 343 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AgentWire CLI Reference
# Session management
agentwire new -s name # not: tmux new-session
agentwire new -s name --no-soul # skip the always-injected soul personality role
agentwire new -s name --first-message "idea" # deliver first prompt once agent boots
# (verified paste, local only; failure ≠ command failure)
agentwire send -s name "prompt" # not: tmux send-keys
agentwire send -s name --wait-ready --timeout 60 -- "prompt"
# wait for agent boot (banner + screen-stable +
# trust-prompt auto-accept), verified delivery,
# exit 1 if unverified; local only
agentwire send-keys -s name key1 key2 # raw keys with pauses
agentwire send-keys -s name --pane 2 key # target a specific pane
agentwire new -s name --created-by orch # FORCE this creator (prompt-routing parent)
# regardless of project; '' forces standalone.
# Default (no flag): inherit the calling tmux session
# ONLY if -p/-s targets the caller's own project — a
# different project defaults to a standalone root (#715)
agentwire output -s name # not: tmux capture-pane
agentwire info -s name # session metadata (cwd, panes) as JSON
agentwire kill -s name # not: tmux kill-session
agentwire list # not: tmux list-sessions
agentwire recreate -s name # destroy and recreate with fresh worktree
# DESTRUCTIVE: rm -rf's the worktree dir + new branch
agentwire restart -s name # relaunch IN PLACE, same conversation (#871) — /exit,
# regenerate flags from the recorded roles/posture/model,
# relaunch at the same cwd with --resume. Nothing on disk
# is touched (unlike recreate) and no new tmux session is
# made (unlike `history resume`). Works on a session that
# ISN'T running — that's the post-reboot/post-rebuild case.
# If the conversation's history is orphaned or gone it
# starts FRESH with the role intact and says so.
# Waits for the agent to come back (exit 1 if it doesn't);
# --no-wait skips. Local only; can't restart itself.
agentwire wait --children # BLOCK on the child sessions you spawned (#852) — collects
# each report, tears the child down, names the ones that
# never reported. Waiting idly instead gets YOU reaped
# mid-fan-out (idle != done). --timeout bounds this call;
# exit 1 = still pending, just call it again.
# Enrollment is automatic on `new`/`worktree` (opt out with
# --no-cohort; --kind orchestrator never enrolls).
agentwire worktree name # new branch + worktree + STANDALONE session
# "worktree session" ALWAYS means this command — never
# `spawn --branch` (that makes a pane). Defaults to the
# bypass posture (autonomous, topology-driven — same for
# every role on this verb); override with --posture.
# ROLE defaults to "worker" (zero behavior change) — the
# worker-worktree etiquette role is intrinsic to that
# default (isolation, no rebuild/restart, verify
# in-worktree, draft PR + notify-back) — first prompts
# only need the task itself
# Base branch (default mode): --base wins, else config
# worktree.default_base, else the repo's actual default
# branch (origin/HEAD, fallback current) — no hardcoded 'main'.
# --project defaults to the git root of cwd (monorepo-safe:
# many sessions can target one repo from different branches).
# config worktree.naming can template the NEW branch name.
agentwire worktree name -b develop # from specific base branch
agentwire worktree name -c # from repo's current branch
agentwire worktree name -e # checkout existing branch (no new branch)
agentwire worktree name --ref v2.0 # detached at tag/commit
agentwire worktree name --prompt "task" # spawn AND seed the first message in one call (verified delivery)
agentwire worktree name --kind orchestrator # ROLE override (#716): a durable, replaceable-
# persona project window instead of a safety-railed
# subordinate. Roots by default (created_by='') unless
# --created-by says otherwise.
agentwire orchestrator [name] -p <project> # sugar for `worktree --kind orchestrator`
# (name defaults to "orchestrator") — the durable-window
# one-liner for a monorepo/large-repo project.
agentwire worktree name --kind reviewer # ROLE override (#827): a PR-review station — safety-
# railed the other way (never opens/merges its own PR,
# iterates via msg_send, reports a verdict via notify_parent).
# Stays parented like worker (not rooted). Pane/main topology
# is the typical default (`agentwire new --kind reviewer`);
# use `worktree --kind reviewer` for a local checkout to e2e
# a sibling's branch.
agentwire worktree --list # list this repo's worktree sessions + read-only git status; --all = every repo
agentwire worktree --status name # read-only git status (dirty/ahead/behind/pushed) for one worktree
agentwire worktree --dangling # LIVE worker sessions with an OPEN PR and no live recorded
# parent (#716) — a PR nobody is positioned to review/merge.
# Distinct from --list's "orphan" (dead session, disk remnant).
agentwire worktree --remove name # ATOMIC teardown: kill session + force-remove worktree + delete
# merged branch (local+remote) + unregister — fails loudly (non-zero,
# registry entry kept) if the dir can't actually be cleared (#717).
# --keep-branch skips branch cleanup; --force-delete-branch deletes
# even if not confirmed merged (but refuses an OPEN PR — #756 — unless
# --close-pr-branch is also given, since that would silently close it).
agentwire worktree --prune # drop registry entries whose worktree is gone + git worktree prune
# --gc-merged: also tear down (session+worktree+branch) any
# still-present entry whose branch is confirmed merged
agentwire tabs track --session name --tab-id <id> [--url <url>] # bookkeeping for a
# claude-in-chrome tab the session opened, so worktree
# teardown can report it if the session never closes it
agentwire tabs untrack --session name --tab-id <id> # drop tracking after closing it yourself
agentwire tabs list [--session name] # list tracked tabs (debug leaked verification tabs)
agentwire fork -s name # fork session into new worktree
agentwire fork -s name -t project/branch --commit abc123 # fork from specific commit
# Helper session — a worker session with NO isolation (#838): shares the
# caller's checkout, so zero git work at creation (no worktree, no branch,
# nothing in the worktree registry). Reproduces a worker pane's one real
# advantage while keeping msg inbox / voice / prompt routing / portal
# visibility. `wait --children` collects its report then reaps it (topology
# "main"), unlike a worktree child which is left alive.
agentwire helper name # worker session sharing this checkout
agentwire helper name -p ~/projects/repo --prompt "run the suite, report failures"
agentwire helper name --roles wiki # extra roles STACK on worker + shared-checkout
# The `shared-checkout` role is auto-injected:
# read/edit freely, NEVER commit/branch/checkout/
# stash/reset/pull — the checkout's owner commits.
# Needs a commit or a PR? Use `worktree` instead.
# Teardown is just `agentwire kill -s name`.
# Pane commands (worker PANES inside the current session — NOT worktree
# sessions; for parallel autonomous work use `agentwire worktree` above)
agentwire spawn --roles worker # spawn worker pane
agentwire spawn --branch name # worker pane on an isolated worktree (still a pane)
agentwire send --pane 1 "task" # send to pane
agentwire output --pane 1 # read pane output
agentwire kill --pane 1 # kill pane
agentwire jump --pane 1 # focus pane
agentwire split -s name # add terminal pane(s)
agentwire detach -s name # move pane to its own session
agentwire resize -s name # resize window to fit largest client
# Boot everything
agentwire up # boot all services (portal, TTS, STT, scheduler,
# custom) then start/attach the dev session
agentwire up --no-tts --no-stt # skip optional voice services
agentwire up --dev # run portal from source (uv run)
# Portal management
agentwire portal start # start in tmux
agentwire portal stop # stop portal
agentwire portal restart # stop + start
agentwire portal status # check health
agentwire portal token # print the bootstrap auth token (host/CLI/MCP credential)
agentwire portal token --rotate # generate a new bootstrap token (re-enter on devices)
# Per-device credentials (#423) — pair a device for its OWN revocable token
agentwire portal pair [--name phone] # print a short-lived pairing code + QR → /pair?code=
agentwire portal devices [--json] # list paired devices (id, name, last-seen, status)
agentwire portal revoke <id> # revoke ONE device (others keep working)
# Scratch pad (shared notes — portal drawer Alt+N; file: ~/.agentwire/scratchpad.json)
agentwire scratchpad list # list notes (newest first)
agentwire scratchpad add "text" --source mysession # add a note (drawer refreshes live)
agentwire scratchpad remove <id> # delete a note
agentwire scratchpad clear # delete all notes
# Custom services (registered long-running sessions — services.custom in config;
# autostart on portal launch, health-checked + restarted by the portal watchdog)
agentwire services list # registry: autostart/restart/healthcheck per service
agentwire services status # run healthchecks now (exit 1 if something's down)
agentwire services status name # one service
agentwire services up <name> # start (also clears 'down' state)
agentwire services up --all # start all autostart services (skips downed)
agentwire services down <name> # stop AND keep stopped (watchdog won't respawn)
# TTS/STT servers
agentwire tts start|stop|status # TTS server management
agentwire stt start|stop|status # STT server management (host shim on :8101)
agentwire stt start # engine from stt.engine (auto|moonshine|whisper); moonshine = fast CPU
agentwire stt start --backend moonshine --model moonshine/base --port 8101 # ad-hoc overrides
# Voice
agentwire say "text" # speak (auto-routes to browser or local)
agentwire say -s name "text" # speak to specific session
agentwire say "spoken" --display "richer card" # speak AND show a desktop toast with different text (one call)
agentwire notify-parent "text" # notify parent session (worker→orchestrator)
agentwire notify-parent --to name "text" # notify specific session
agentwire notify-parent --raw --to name "text" # verbatim, no [NOTIFY ...] prefix
# (delivery is safety-gated: refuses targets showing a
# live dialog / bare shells / parked sessions, verified paste)
# Prompt routing (interactive prompts → parent session; see wiki sessions/prompt-routing.md)
agentwire prompts status # pending prompt markers
agentwire prompts tick # run one sweep now (watchdog does this every 60s)
agentwire prompts answer -s name --pane 0 --expect <hash> 2 # guarded answer:
# re-detects + hash-compares before sending keys —
# NEVER answer dialogs with raw send-keys
agentwire prompts clear -s name --pane 1 # drop a marker
# Polite messaging (non-interrupting agent-to-agent inbox; see wiki sessions/messaging.md)
agentwire msg send --to name "text" # queue a message (delivers when their box is clear)
agentwire msg send --to name --kind done "PR #312 drafted" # kinds: note|done|request|escalation|ingest|voice|idle (idle = the idle hook's synthetic placeholder, #952)
agentwire msg send --to name --kind ingest --ref "/path/report.md" "topic" # PASSIVE: never auto-delivered, pull-only
agentwire msg send --to @all "team update" # broadcast to live agent sessions except sender
agentwire msg send --to name --body-file /tmp/body.md # code-bearing body, no shell escaping ('-' = stdin)
agentwire msg inbox -s name # peek pending + passive (does not drain/consume)
agentwire msg pull -s name # read + REMOVE passive (ingest) messages — the voluntary pull
agentwire msg dead -s name # list dropped (dead-lettered) msgs + reason/timestamp
agentwire msg flush -s name # attempt a drain now (still gated on empty box + safe target)
# Research dropbox (Briefing Mode)
agentwire research dir -s name # print the dropbox path (~/.agentwire/research/<session>/)
agentwire research ensure -s name # create + print the dropbox path
# `msg` NEVER clobbers a human's draft — unlike `send`, which
# pastes + Enter immediately. Use `send` only to forcibly drive a session now.
agentwire listen start|stop|cancel # host voice recording (needs stt.backend: custom + the :8101 shim)
agentwire listen stop -s name # transcribe + send to a tmux session (default)
agentwire listen stop --type # transcribe + type at cursor (Hammerspoon paste)
agentwire listen stop --stdout # transcribe + print raw transcript to stdout, no paste/send
# (scripting hook — Hammerspoon etc. capture $())
# Voices (custom/cloned voices come from a TTS shim — see docs/wiki/voice/shim-contract.md)
agentwire tts voices # list available voices (custom-shim voices or Kokoro presets)
# Artifact windows (agent visual canvas)
agentwire open <url> --title "T" # announce URL/local file as a click-to-open artifact notification (#817)
agentwire open dashboard.html # announce from ~/.agentwire/artifacts/ — human clicks to open
# Channels (outbound notification integrations — email + quo)
agentwire channels list # list all registered channels
agentwire channels list --json # JSON output
# Email (send-only channel)
agentwire email --to addr --subject "Subject" --body "Body"
agentwire email --body "msg" # uses default_to from config
agentwire email --attach file.pdf --body "See attached"
# Quo SMS (send-only channel, no deps)
agentwire quo --body "msg" --to "+1234567890"
# Machine management
agentwire machine list
agentwire machine add <id> --host <host> --user <user>
agentwire machine remove <id>
# SSH tunnels (for remote services)
agentwire tunnels up # create all required tunnels
agentwire tunnels down # tear down all tunnels
agentwire tunnels status # show tunnel health
agentwire tunnels check # verify tunnels are working
# Lock management (for scheduled tasks)
agentwire lock list # list all locks
agentwire lock clean # remove stale locks
agentwire lock remove <session> # force-remove a specific lock
# Project discovery
agentwire projects list # discover projects from projects_dir
agentwire projects list --json # JSON output for scripting
agentwire projects create name # mkdir + minimal .agentwire.yml (bypass)
# (in git repos, .agentwire.yml is auto-added to
# .gitignore — personal config, keep it untracked)
agentwire projects create name --git-init # also run `git init`
agentwire projects create name --from URL # git clone URL instead of mkdir
# Session history
agentwire history list # list conversation history
agentwire history show <id> # show session details
agentwire history resume <id> # resume session (always forks)
# Shareable conversation handoffs (issue #157)
agentwire handoff init [--title hint] # create bundle dir + pre-filled ai-handoff.md template
agentwire handoff render <bundle-dir> # render show-the-story.html from ai-handoff.md
agentwire handoff list # list past bundles
# Inside a Claude Code session, prefer the /handoff slash command — it walks the
# agent through filling the template using full conversation context (free, no
# fresh LLM call). Outputs land in ~/.agentwire/artifacts/handoff-<slug>/.
# Roles management
agentwire roles list # list available roles
agentwire roles show <name> # show role details
# Scheduled workloads
agentwire ensure -s name --task task # run named task reliably
agentwire task list [session] # list tasks for session/project
agentwire task show session/task # show task definition
agentwire task validate session/task # validate task syntax
# URL fetch (helper usable from any session, including pi)
agentwire fetch <url> # fetch a page via Jina Reader (markdown, JS-rendered)
agentwire fetch <url> --limit 4000 # cap chars (default 8000, 0 = no limit)
# Safety & diagnostics
agentwire safety check "cmd" # test if command would be blocked
agentwire safety status # show pattern counts and recent blocks
agentwire safety logs # query audit logs
agentwire safety install # install damage control hooks
agentwire hooks install # install permission hook (Claude Code only)
agentwire hooks uninstall # remove permission hook (Claude Code only)
agentwire hooks status # check hook installation status
agentwire network status # complete network health check
agentwire doctor # auto-diagnose and fix issues
agentwire doctor --voice # only the push-to-talk path: mic, STT shim, portal/tunnel, tmux+PTT (pass/fail + fix)
# Notifications (the notify-* family, by target)
agentwire notify-event EVENT # broadcast a portal lifecycle event (session/pane); usually called by tmux hooks
agentwire notify-parent "text" # text up to your parent/orchestrator session
agentwire notify-user "text" # desktop toast for the human (safe markdown: bold, [links](url), line breaks)
# MCP Server
agentwire mcp # expose agentwire as MCP server
# Scheduler
agentwire scheduler start|serve|stop|status # manage scheduler daemon
agentwire scheduler board # show task board with overdue scores
agentwire scheduler live # show live scheduler state
agentwire scheduler events # show recent scheduler events
agentwire scheduler history # show recent run history
agentwire scheduler run task # force-run a task now
agentwire scheduler enable|disable task # enable/disable a task
agentwire scheduler report [--since 8h] [--artifact] # generate morning report HTML
agentwire scheduler dashboard # open scheduler dashboard
# Usage-limit recovery (deterministic watchdog, see docs/wiki/usage-limit-recovery.md)
agentwire limits tick # one watchdog pass: sweep panes, resume what's due
agentwire limits status # show sessions parked on usage limits
agentwire limits resume -s name [--force] # manually resume a parked session now
agentwire limits install # install + load the launchd watchdog (60s tick)
agentwire limits uninstall # unload + remove the watchdog
# Setup & Development
agentwire init # interactive setup wizard (ends on the portal URL)
agentwire init --assisted # ...and spawn the Claude TTS/STT setup session at the end
agentwire generate-certs # generate SSL certificates
agentwire up # boot all services + dev session (see "Boot everything")
agentwire dev # start/attach to dev session ONLY (no services)
agentwire rebuild # clear uv cache and reinstall
agentwire uninstall # uninstall the tool
Changes
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 343 lines · 88 tokens per session scan C 44078ff4442d
agentwire-cli is a skill published in the GitHub repository dotdevdotdev/agentwire-dev (21 stars, last pushed 17d ago), licensed Apache-2.0. It adds 88 tokens to every session and 4,947 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Related
Other skills, from other repositories
luvus-module
Write a luvus module (an extension for luvus, mission control for your AI coding agents). Use when the user is building or debugging a luvus module: authoring luvus-module.toml, adding a sidebar dock, Luvus Bar widget, right-click action, event hook, module pane, module settings, or calling luvus back over its UHP.
luvus
Control Luvus through its local CLI and UHP. Use only for a line beginning with =target message, an explicit request naming Luvus, a request to delegate to a named live Luvus agent or pane, or an explicit Luvus operation involving sessions, workspaces, tabs, panes, agents, files, Git, DIFF, worktrees, tasks, leases…
tmux-via-mcp
Use the tmux MCP tools to create sessions, shape layouts, run tracked commands, and automate interactive terminals when a real TTY or parallel panes are required.
tmux-buffer-explorer
Explore large tmux buffers via search and bounded slices. Use when buffer data is too large to load at once or needs incremental inspection.
not-happy-jan
Adjust Not-Happy-Jan feedback settings by running the nhj CLI when the user wants to tune voices, dials, muting, censoring, tests, or hold music.
release
Use when the user asks to release unitmux, bump the version, publish a GitHub release, update the Homebrew cask, check release status, or says リリース/release/バージョン上げて.