SkillAgentSearch skills...

open-agent-teams

Delegate tasks to ANY CLI agent (claude, codex, aider, ...) running in a detached tmux session, with a race-safe done-signal protocol and multi-turn iteration

Install / Use

npx skills add AI-Builder-Club/skills --skill open-agent-teams

Installs into whichever agent you are using.

About this skill
📄

SKILL.md

Installable skill definition

Quality Score

83/100

Supported Platforms

Claude Code
OpenAI Codex
Aider

Tags

Our assessment of open-agent-teams

open-agent-teams scores 83/100 on our quality scale, 3154th of 4,698 Development & Engineering skills we index.

Its SKILL.md is 5.4 KB long, split into 5 sections with 1 code example: a solid amount of guidance for an agent.

With 1,274 GitHub stars, it is one of the more widely adopted skills in the catalogue.

Substance
26/30
Structure
15/20
Description
15/15
Adoption
13/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated about 2 months ago, so open-agent-teams is actively maintained.
  • Our last check on 2026-11-07 found the source still online.
  • No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
  • Its trust signals score 88/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

Review

Our scan of the whole file found 3 patterns worth reviewing before you install open-agent-teams. An AI review judged it risky: Line 25 gives an example using `claude --dangerously-skip-permissions` to start a delegated agent.

  • mediumDisables the agent's permission promptsline 25
    # e.g. tdel start hello "claude --dangerously-skip-permissions" "make a hello world page"
  • mediumDisables the agent's permission promptsline 52
    | claude | `claude --dangerously-skip-permissions` (`--model`, `--effort low..max`) | `esc to interrupt` | `/exit` | Escape |
  • mediumDisables the agent's permission promptsline 53
    | codex | `codex --dangerously-bypass-approvals-and-sandbox` (`--model`, `-c 'model_reasoning_effort="low..xhigh"'`) | `esc to interrupt` | …

AI review: risky

  • Line 25 gives an example using `claude --dangerously-skip-permissions` to start a delegated agent.
  • Lines 52-53 list harness commands `claude --dangerously-skip-permissions` and `codex --dangerously-bypass-approvals-and-sandbox` as recommended ways to launch agents.
  • The harness also recommends `grok --always-approve` and notes `pi` has no permission system, collectively instructing the agent to disable or bypass permission/approval safeguards.

AI review by kimi-k2.7-code on 2026-10-04. Automated pattern scan on 2026-10-04. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

open-agent-teams compared with similar skills

All 4 of these similar skills score higher than open-agent-teams; compare them before choosing.

SkillScoreStarsUpdatedFormat
open-agent-teams (this skill)by AI-Builder-Club831.3k2mo agoSKILL.md
taste-skillby Leonxlnx100116.8k8d agoCLAUDE.md
ai-job-searchby ai-job-search-dev10072.1ktodayCLAUDE.md
claude-howtoby luongnv8910047.3k2d agoCLAUDE.md
lobehubby lobehub10085.6ktodayMCP Server

Frequently asked questions

How do I install open-agent-teams?
Run npx skills add AI-Builder-Club/skills --skill open-agent-teams. The install tabs above show the steps for each supported agent.
Which AI agents does open-agent-teams work with?
It is written for Claude Code, OpenAI Codex and Aider, as a SKILL.md file. Other agents that read the same format can often use it too.
Is open-agent-teams safe to use?
Our scan of the whole file found 3 patterns worth reviewing before you install open-agent-teams. An AI review judged it risky: Line 25 gives an example using claude --dangerously-skip-permissions to start a delegated agent. It declares no license and scores 88/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is open-agent-teams still maintained?
The repository was last updated about 2 months ago, so open-agent-teams is actively maintained.

name: open-agent-teams description: Delegate tasks to ANY CLI agent (claude, codex, aider, ...) running in a detached tmux session, with a race-safe done-signal protocol and multi-turn iteration. Use when delegating work to a non-Claude CLI agent, when the user says "tmux delegate", "run agent in tmux", "delegate to codex/aider", or when executor work should run in an observable background terminal instead of the Agent tool.

open-agent-teams

Run any CLI agent as an executor inside a detached tmux session. Communication is: prompt in via CLI arg / send-keys, completion out via a file sentinel the agent touches, summary out via a result file the agent writes. File sentinels are used instead of tmux wait-for because a wait-for -S with no waiter is silently lost; files never race and allow timeouts.

All commands via the helper (make sure it runs from the skill dir):

SKILL_DIR/scripts/tdel

Initial setup (once per repo)

If the repo's CLAUDE.md (or AGENT.md) has no delegation rules yet, add the contents of references/CLAUDE.delegation-template.md to it — it defines the coordinator/executor roles and the ROLE: EXECUTOR prompt marker used below.

Workflow

  1. Start an agent on a task:

    tdel start <session> "<agent-cmd>" "<prompt>"
    # e.g. tdel start hello "claude --dangerously-skip-permissions" "make a hello world page"
    

    This appends the done-protocol to the prompt automatically (write summary to result file, then touch the done file).

  2. Wait for completion — ALWAYS via a background Bash task so you get woken up instead of blocking or polling:

    tdel wait <session> [timeout-sec]     # run with run_in_background: true
    

    Exit 0 = done (prints the agent's summary). Exit 124 = timeout (prints last pane lines for diagnosis). Non-zero also if the session died.

  3. Review the output (tdel result <session>, plus inspect the actual files the agent changed). If iteration is needed:

    tdel send <session> "<feedback / next instruction>"
    tdel wait <session>                   # again, run_in_background
    

    Each send is a new turn with its own done/result files — no signal cross-talk between turns.

  4. Debug a stuck or slow agent: tdel peek <session> [lines] shows the live pane. tdel status lists all sessions.

  5. Stop when finished: tdel stop <session> (kills the session and removes state under /tmp/agent-delegate/<session>).

Harness reference

How to launch each CLI agent autonomously (prompt is passed as the last positional arg by tdel start):

| Harness | agent-cmd for tdel start | Busy-pane signature | Exit | Interrupt | |---|---|---|---|---| | claude | claude --dangerously-skip-permissions (--model, --effort low..max) | esc to interrupt | /exit | Escape | | codex | codex --dangerously-bypass-approvals-and-sandbox (--model, -c 'model_reasoning_effort="low..xhigh"') | esc to interrupt | /quit | Escape | | grok | grok --always-approve (--model, --reasoning-effort low\|medium\|high) | Ctrl+c:cancel | Ctrl+Q twice within 1s (NOT /exit, NOT Ctrl+C) | Ctrl+C (Escape does NOT interrupt) | | pi | pi (--model, --thinking low..max; no permission system — always autonomous) | Working... | /quit | Escape | | opencode | opencode --prompt-style launch (--model provider/model) | esc interrupt (no "to") | /exit | double Escape (flaky mid-shell-command) |

Busy signatures matter on timeout: peek and grep for the signature — present = still working (extend the wait), absent = idle (it likely finished without touching the done file, or is stuck at a prompt).

Trust dialogs (first run per repo/worktree): claude (trust/bypass-permissions confirm), codex ("Do you trust..."), and pi can each show a dialog that blocks the prompt from processing. After start, peek within ~20s; if a dialog is showing, accept it with tdel key <session> Enter and verify via another peek that the task started. grok skips its picker when launched inside a git repo root.

Slash/skill popup hazard: prompts beginning with / or $ (skill invocations: /<skill> on claude/grok, $<skill> on codex) open an autocomplete popup — a fast Enter selects the popup instead of submitting, and grok needs a genuine second Enter. tdel send handles this automatically (longer settle + double Enter for /- or $-prefixed prompts).

Resume after exit: codex codex resume <session-id>, grok grok --resume <session-id> (ids printed on quit), opencode relaunch with --continue.

Rules

  • Prompts must be self-contained (context, constraints, expected output) — the executor can't see your conversation. Prefix with the repo's executor marker (ROLE: EXECUTOR ...) when the agent is Claude Code in a repo using the delegation rules (see Initial setup above).
  • Prompts are flattened to one line before sending (newlines would submit a TUI input box early) — write them accordingly; put large specs in a file and reference the path.
  • On timeout, peek first — agents sometimes finish but forget to touch the done file. If the work is visibly done, treat pane output + changed files as the result and stop or send a reminder.
  • Don't trust capture-pane as the deliverable — it's a rendered TUI snapshot. The result file and the actual changed files are the source of truth.
  • One task per session; parallel tasks = parallel sessions with distinct names.

Related Skills

View on GitHub
GitHub Stars1.3k
CategoryDevelopment
Updated2mo ago
Forks158

Languages

Python

Trust signals

88/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 medium