New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Agent Bridge

by FeiZhuLulu in FeiZhuLulu/Agent-Bridge

Coordinate local worker agents (Grok Build, Kimi Code, Antigravity, DeepSeek Harness, OpenCode, Claude Code, Codex CLI, Devin CLI, ZCode, MiniMax Code) through the Agent Bridge MCP tools.

MITAuto-check passedAgent Workflows

Install Agent Bridge

skills CLI
$ npx skills add FeiZhuLulu/Agent-Bridge --skill agent-bridge -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install FeiZhuLulu/Agent-Bridge agent-bridge --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/FeiZhuLulu/Agent-Bridge.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/agent-bridge .claude/skills/agent-bridge && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
agent-bridge
GitHub stars
110
Token cost
~2.4k tokens
SKILL.md length
1,269 words
Files
1
Skills in repo
3
Repo updated
First seen
Licence
MIT

At a glance

Coordinate local worker agents (Grok Build, Kimi Code, Antigravity, DeepSeek Harness, OpenCode, Claude Code, Codex CLI, Devin CLI, ZCode, MiniMax Code) through the Agent Bridge MCP tools.

  • Works in 2 steps: dispatch, or do it yourself? → which worker (user instructions override…
  • Dispatching implementation
  • SKILL.md covers Before anything: read the policy, Step 1 — dispatch, or do it…, Step 2 — which worker (user… and The dispatch loop
  • Calls git and codex

What it does

Agent Bridge is an agent skill from FeiZhuLulu/Agent-Bridge. Coordinate local worker agents (Grok Build, Kimi Code, Antigravity, DeepSeek Harness, OpenCode, Claude Code, Codex CLI, Devin CLI, ZCode, MiniMax Code) through the Agent Bridge MCP tools. Use when dispatching implementation, research, or test work to a worker; when the user mentions Agent Bridge, dispatchtask, or a worker by name; or when deciding whether to delegate a coding task instead of doing it yourself.

Its SKILL.md is about 2.4k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in Agent Workflows, covering MCP servers. It works with Kimi, DeepSeek and MiniMax. The repository describes itself as: Agent Bridge is a connector that links various local agents, enabling users to coordinate multiple other agents simply by interacting with a single powerful agent. It currently… The licence is MIT.

When your agent uses it

  • Dispatching implementation
  • Test work to a worker
  • The user mentions Agent Bridge
  • A worker by name

Example prompts

  • “/agent-bridge”

Workflow steps

2 steps, taken from the step headings in SKILL.md.

  1. dispatch, or do it yourself?
  2. which worker (user instructions override this)

What it can do on your machine

Read from SKILL.md and the folder at commit 5560e06. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • git
    • codex

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use git, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Agent Bridge loads about 2.4k tokens when it runs. Until then it costs about 107 tokens; SKILL.md has 1,269 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~107
When it runs · the whole SKILL.md, loaded when a task matches
~2.4k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from FeiZhuLulu/Agent-Bridge at commit 5560e06, republished under its MIT licence (© FeiZhuLulu). 1,269 words, ~2,403 tokens.

Download SKILL.mdSave it as .claude/skills/agent-bridge/SKILL.md (or your agent's skills folder).
name
agent-bridge
description
Coordinate local worker agents (Grok Build, Kimi Code, Antigravity, DeepSeek Harness, OpenCode, Claude Code, Codex CLI, Devin CLI, ZCode, MiniMax Code) through the Agent Bridge MCP tools. Use when dispatching implementation, research, or test work to a worker; when the user mentions Agent Bridge, dispatch_task, or a worker by name; or when deciding whether to delegate a coding task instead of doing it yourself.

Dispatching workers through Agent Bridge

You are the coordinator: users talk only to you, and you call the workers. Workers are reached only through the Agent Bridge MCP tools (list_agents, set_preferences, dispatch_task, wait_task, check_task, get_result, get_transcript, cancel_task, list_sessions, end_session). If those tools are missing from this session, stop and say so — never run grok, kimi, agy, dsh, opencode, claude, claude-agent-acp, codex, devin, zcode-acp-server, or mcode CLIs directly, and never drive their GUIs. The same product can be a coordinator and a worker; those are different processes.

Before anything: read the policy

Call list_agents first. Its result carries:

  • coordinator.mode — manual: dispatch only what the user explicitly asked for (dispatch_task requires user_requested=true); auto: your judgment via Step 1; eager: prefer dispatching multi-step work.
  • coordinator.instructions — the user's persistent routing preferences. They override Step 2 below. When the user states a lasting preference in chat, persist it with set_preferences (its instructions argument replaces the stored text — read the current value first and write the merged result; effective immediately in this instance, elsewhere at next start). One-off wishes are not preferences; follow them without persisting.
  • coordinator.runtime_context / coordinator.dispatch_enabled — a top-level host is coordinator / true. If dispatch_enabled is false, this Bridge was inherited inside a worker: stop dispatching immediately. Do not call dispatch_task, set_preferences, cancel_task, or end_session. user_requested=true does not bypass that. Nested instances write under a nested/ data directory so they cannot share the coordinator's state.json.
  • env.proxy / env.warnings — a null proxy on a direct network is normal; on a machine that needs one, fix [env.proxy] in agents.toml instead of retrying a failed dispatch.
  • agents[].quota — remaining plan per worker: status ok / exhausted / unknown, windows[] with remaining_percent and resets_at / resets_in_sec (Codex 5h + weekly, Kimi Code limits, Claude 5h + weekly, Grok weekly), optional balance when reported by the provider, plan, cached / stale, detail. Information, not a routing rule: exhausted means the turn will most likely fail — prefer another worker or tell the user when it resets; unknown means Bridge could not read it (unsupported CLI, API-key login, timeout — the reason is in detail), never that it is empty. Quota does not change available. Custom API/auth endpoints are unsupported and return unknown before credentials or cached quota are read. Cached readings expire at the earliest window reset. DSH balance lookup is unsupported.

Claude status: shared 5h/weekly only. Before dispatch, even if ok, check the target model's weekly:opus/weekly:sonnet: 0% is exhausted; missing/null unknown. Report reset time or a user-permitted alternative. Model-only data leaves shared status unknown.

Step 1 — dispatch, or do it yourself?

A cost question, not a category question; "it is implementation work" is never by itself a reason to dispatch.

Do it yourself when: after reading 1–2 files you already know the exact edit; the whole job is reading a little code and answering; or writing the dispatch message would cost more than making the change.

Dispatch when: the change spans several files or needs exploration you have not done; tests must be written or a build/test loop iterated; breadth research across many sources; or not dispatching would eat many turns of mechanical work.

Examples: "fix the README typo" → yourself. "Add a None check at line 120" → yourself, even though it is implementation. "Add retry logic to the ACP adapter with tests" → Grok. "Port this 4000-line module" → Kimi (kimi-code/k3-256k). "Survey how other CLIs handle session resume" → Antigravity.

Step 2 — which worker (user instructions override this)

  • Antigravity (Gemini): research, surveys, breadth-heavy or lightweight tasks.
  • Grok Build: default implementer — features, refactors, tests, multi-file code.
  • Kimi Code: second implementer — Grok busy or wrong, independent second take, or big single-context jobs.
  • OpenCode: optional third implementer — user asked for it, wants a connected provider/model, or Grok and Kimi are busy.
  • Claude Code: optional implementer — user asked, or Grok and Kimi are busy. Worker binary is claude-agent-acp.
  • Codex CLI: optional implementer — user asked, or others are busy. Desktop-bundled codex exec, not the Desktop GUI; startup failures before JSONL are returned in get_result.error.
  • Devin CLI: optional implementer — user asked, or others are busy. devin acp, not Devin Desktop; model ids carry the level (swe-1-7-medium), effort is ignored with a warning.
  • ZCode: optional implementer — user asked, or others are busy. Worker is zcode-acp-server, not the ZCode app. Mode is forced to yolo.
  • MiniMax Code: optional implementer — user asked, or others are busy. mcode acp. Permission is forced to bypassPermissions.
  • DeepSeek Harness: only when others are unavailable or the user asks.
Show full SKILL.md (546 more words)Show less

The dispatch loop

  1. dispatch_task with cwd = this conversation's project folder (absolute) — never the Agent Bridge install path, never a temp dir. The message must be self-contained: background, absolute paths, acceptance criteria, things not to do. Leave model/effort unset unless you have a reason; slugs and effort mappings per worker are documented in ORCHESTRATION.md in the Agent-Bridge repo.
  2. Loop wait_task until terminal. A timeout is not failure — call it again. silent_for_sec in the payload is the time since the worker's last output; a turn silent past stall_timeout_sec (default 1800) ends failed / stop_reason="stalled" — raise that worker's limit in agents.toml if the step was legitimately long. Stall counts only worker output; turn_timeout_sec (default 0 = off) instead caps a turn's wall-clock and ends failed / timed_out. dispatch_task and every terminal task answer (wait_task / check_task / get_result / cancel_task) carry persisted; false means state.json could not be written — the answer is still valid but may not survive a restart. files_changed lists only paths inside the task cwd; outside-cwd paths land in files_outside_cwd (absolute, capped the same way, files_outside_cwd_total). A warning that session/load failed and prior context is lost means a fresh native session — re-supply context next message. Size timeout_sec under the host MCP tool timeout:
    • Codex: tool_timeout_sec 600; default 180 is fine.
    • Cursor: host ~45–60 s; pass ~30 and loop.
    • Kimi Code: configure toolTimeoutMs 600000; otherwise ~45 s polls.
    • ZCode: configure timeoutMs 600000; otherwise ~15–20 s polls.
    • Grok Build: official default tool_timeout_sec is 6000; set 600. If unsure or the host kills the call, ~30–45 s polls.
    • Claude Code: per-server timeout 600000 (ms) in .mcp.json. CLI default is long; if unsure, ~45 s polls.
  3. get_result; while has_more is true, call it again with cursor=next_cursor and concatenate the pages. Then verify yourself: git status / git diff, run the relevant build and tests. Never trust the worker's self-report. Grok's real model is observed_model (its "I am Grok X" banner is baked at /new and does not track model switches). OpenCode, Claude Code, ZCode, and MiniMax Code observed_model / observed_effort are the last values Bridge set after mapping, not a live sampler. An empty Kimi result with non-empty warnings is a failed turn, not a no-op.
  4. If review fails, dispatch_task again on the same session_id with a concrete problem list — at most three follow-ups, then fix it yourself and tell the user.
  5. Summarize the diff, leftover risk, and worker usage. end_session when the worker is no longer needed.

ACP failures retain string error plus optional failure. If failure.prompt_may_have_been_sent is true or evidence is missing, do not blindly replay; false is not proof of safe retry. Bridge never auto-retries prompts. See SETUP.md for diagnostic fields.

For optional retry deduplication, send a UUID request_id on the first dispatch_task call, then retry with the same ID and original arguments; keep session_id omitted if it was originally omitted. Adding an ID only on retry cannot deduplicate the first call. Identical retries reuse the task (reused=true) only within the same Bridge instance while the task is retained. Different arguments are rejected, and normal dispatch validation still applies. Restart, another instance, or task pruning loses the binding; this does not guarantee exactly-once worker side effects.

check_task / get_result on a task owned by another Bridge instance answer from state.json (from_state: true, silent_for_sec: null); wait_task / cancel_task still require the owning instance.

© FeiZhuLulu, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in skills/agent-bridge of FeiZhuLulu/Agent-Bridge.

Open the folder on GitHubat commit 5560e06

Compare with similar skills

Agent Bridge next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Agent Bridge compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Agent Bridge this skillFeiZhuLulu/Agent-Bridge110—~2.4kAutomated safety check: PassMIT
Subagentethanhq/cc-fleet216—~4.9kAutomated safety check: PassApache-2.0
Subagentethanhq/cc-fleet216—~4.3kAutomated safety check: PassApache-2.0
Claude Switch Models Setupdaymade/claude-code-skills1.4k—~11kAutomated safety check: PassMIT
LLM Council on Fireworks AIdair-ai/dair-academy-plugins614—~5kAutomated safety check: NotesMIT
Update Ollama Cloud Modelsheypinchy/pinchy182—~3.9kAutomated safety check: NotesAGPL-3.0

Similar skills

  • Subagent

    ethanhq/cc-fleet

    Fan out a one-shot or flat parallel batch of cc-fleet PROVIDER subagents (headless cc-fleet subagent) that return a result — DeepSeek / GLM / Kimi / Qwen / MiniMax, or a Codex/Claude subscription.

    216 GitHub stars~4.9k tokensUpdated 11 days ago
    Agent WorkflowsAuto-check passed

  • Subagent

    ethanhq/cc-fleet

    Run a one-shot or flat parallel batch of provider LLM subagents (headless cc-fleet subagent) that return a result.

    216 GitHub stars~4.3k tokensUpdated 11 days ago
    Agent WorkflowsAuto-check passed

  • Claude Switch Models Setup

    daymade/claude-code-skills

    Sets up and maintains isolated Claude Code CLI profiles to run Kimi, MiniMax, DeepSeek, GLM, StepFun or Anthropic in separate terminal windows.

    1.4k GitHub stars~11k tokensUpdated today
    Agent WorkflowsAuto-check passed

  • LLM Council on Fireworks AI

    dair-ai/dair-academy-plugins

    Has several open-weight models answer a question, rank each other's anonymized answers, then lets a chairman model write the final response through Fireworks AI.

    614 GitHub stars~5k tokensUpdated 2 mo ago
    AI & LLM EngineeringAuto-check: notes

  • Update Ollama Cloud Models

    heypinchy/pinchy

    A skill your agent uses when a new Ollama Cloud model is announced or available (e.g.

    182 GitHub stars~3.9k tokensUpdated 20 days ago
    AI & LLM EngineeringAuto-check: notes

  • OmniRoute Chat CLI

    diegosouzapw/OmniRoute

    Sends chat completions, streams responses, and opens an interactive REPL against any OmniRoute-routed model provider.

    75k GitHub stars~345 tokensUpdated today
    AI & LLM EngineeringAuto-check passed

More from FeiZhuLulu/Agent-Bridge

  • Add Worker

    FeiZhuLulu/Agent-Bridge

    Connect a new worker agent CLI to Agent Bridge. An agent skill from FeiZhuLulu/Agent-Bridge.

    110 GitHub stars~2.9k tokensUpdated 4 days ago
    Auto-check passed

  • Add Coordinator

    FeiZhuLulu/Agent-Bridge

    Verify that a new host can act as an Agent Bridge coordinator over stdio MCP.

    110 GitHub stars~1.9k tokensUpdated 4 days ago
    Auto-check passed

Works with

Categories

Questions about Agent Bridge

What does Agent Bridge do?

Coordinate local worker agents (Grok Build, Kimi Code, Antigravity, DeepSeek Harness, OpenCode, Claude Code, Codex CLI, Devin CLI, ZCode, MiniMax Code) through the Agent Bridge MCP tools. Agent Bridge is an agent skill from FeiZhuLulu/Agent-Bridge. Coordinate local worker agents (Grok Build, Kimi Code, Antigravity, DeepSeek Harness, OpenCode, Claude Code, Codex CLI, Devin CLI, ZCode, MiniMax Code) through the Agent Bridge MCP tools.

When should I use Agent Bridge?

Agent Bridge fits situations like: dispatching implementation; test work to a worker; the user mentions Agent Bridge; A worker by name.

How do I install Agent Bridge in Claude Code?

Run `npx skills add FeiZhuLulu/Agent-Bridge --skill agent-bridge -a claude-code`. Or copy the skill folder (skills/agent-bridge in FeiZhuLulu/Agent-Bridge) into .claude/skills/agent-bridge in your project. Claude Code loads it when a task matches its description.

How do I install Agent Bridge in Codex?

Run `npx skills add FeiZhuLulu/Agent-Bridge --skill agent-bridge -a codex`. Or copy the skill folder (skills/agent-bridge in FeiZhuLulu/Agent-Bridge) into .agents/skills/agent-bridge in your project. Codex loads it when a task matches its description.

Can I use Agent Bridge in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add FeiZhuLulu/Agent-Bridge --skill agent-bridge -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/agent-bridge, .gemini/skills/agent-bridge, .github/skills/agent-bridge and .opencode/skills/agent-bridge in your project.

What does Agent Bridge need to run?

Going by SKILL.md and its folder, Agent Bridge needs the command-line tools its instructions call (git and codex).

Does Agent Bridge access the network?

SKILL.md contains no URLs. Its commands use git, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is Agent Bridge safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Agent Bridge use?

Agent Bridge is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Agent Bridge use?

About 2.4k tokens (SKILL.md is roughly 9.6k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Agent Bridge?

Skills that share tags, products or a category with Agent Bridge: Subagent (ethanhq/cc-fleet, 216 stars), Subagent (ethanhq/cc-fleet, 216 stars), Claude Switch Models Setup (daymade/claude-code-skills, 1.4k stars) and LLM Council on Fireworks AI (dair-ai/dair-academy-plugins, 614 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Agent Bridge?

FeiZhuLulu (a GitHub user) maintains it in FeiZhuLulu/Agent-Bridge, which has 110 GitHub stars. The repository holds 3 skills in this directory. The repository was last updated on October 7, 2026.

Source: FeiZhuLulu/Agent-Bridge on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.