claude

work-orchestrator

Opinionated orchestration of work through two delegation surfaces: dispatched workers (Codex CLI (GPT-6) and Claude subagents) via a task-strategy workflow with model and reasoning-effort routing, fenced write scopes, and independent two-worker verification; and sibling Claude Code sessions via a distribution workflow with session enumeration, mandatory dispatch-message blocks, and explicit stand-down closure. Ships 11 duty agent definitions that pin no model or effort; the routing table sets model and reasoning effort per checkpoint type at spawn time. Both skills are extendable at five workflow positions and via recognized named configuration values; the plugin ships hooks that deliver a project's per-skill extension files (.claude/extensions/work-orchestrator/) wrapped in a structural envelope.

11 on GitHubby Martin Bensv5.1.1MITupdated 7 days agodevelopment
/

Plugin work-orchestrator from marketplace itb-ai-tools in it-bens/ai-tools. Commands read from the repo's manifest — 6 days ago.

This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.

What it installs

Skills (2)

  • orchestrating-session-work

    Use when work will be distributed to sibling Claude Code sessions, before the first cross-session message. Takes the work items plus the session topology; returns a stated distribution strategy, dispatched handoffs, and per-item adjudicated closure with an explicit stand-down to every engaged sessio

  • orchestrating-subagent-work

    Use for substantial implementation or review work — multi-file changes, review rounds, verification passes — including when the work arrives as an assignment message from another session; invoke before the first codex dispatch, subagent spawn, or inline write. Takes the task plus the project's gates

Subagents (11)

  • adversarial-check

    Use when a dispatch supplies a claim and the artifact it concerns, and deliberately withholds the reasoning of whoever produced it. Returns a verdict from the caller's vocabulary with the source evidence for it, defaulting to refuted where the evidence is not there. Checks adversarially. Does not im

  • design

    Use when a dispatch supplies a requirement, its constraints, and the code to design against, and the caller wants an approach reached without its own reasoning in the way. Returns an approach with sequencing, named trade-offs, and every decision it had to make with the reason. Works from fresh conte

  • escalated-check

    Use when a dispatch supplies a claim and the artifact it concerns and a cheaper check came back thin, unconvincing, or contradicted by something the checker noticed but could not settle. Returns a verdict with its evidence and what would falsify it. Dispatched on a stronger model, or on the same mod

  • extract

    Use when a dispatch supplies an enumeration or extraction task with a stated output schema and no decisions left open. Returns the schema fully populated plus an explicit list of items it could not classify. Sustains high-volume mechanical passes. Does not infer intent, resolve ambiguity, or choose

  • gate-run

    Use when a dispatch supplies exact commands and the directory to run them in, and the caller needs the raw result rather than a verdict on it. Returns per command the command as run, its exit code, and the verbatim tail of its output. Executes and transcribes. Does not interpret results, decide pass

  • implement

    Use when a dispatch supplies a fenced file list, a decided design per item, test duties, and gate commands. Returns a verdict per item — status, files touched, deviations — plus an honest not-verified list, anything the designs did not account for (an interaction between items, a fact the surroundin

  • lookup

    Use when a dispatch supplies one closed question together with the place to look for its answer. Returns the answer with the file and line supporting it, or a not-found that names the scope searched. Answers literally within a fixed scope. Does not widen the scope or answer an adjacent question it n

  • root-cause

    Use when a dispatch supplies an observed behavior and the code path suspected of producing it, and the caller needs the mechanism rather than a remedy. Returns the mechanism, the evidence chain that establishes it, and the confidence in each link. Reasons about causes rather than symptoms. Does not

  • search

    Use when a dispatch supplies a target described by name, pattern, symbol, or convention plus the breadth to cover, and needs the locations back rather than a judgement about them. Returns absolute paths with the excerpt that identified each match, or an explicit empty result naming the scope searche

  • source-conflict

    Use when a dispatch supplies two or more sources that disagree and states what turns on the disagreement. Returns which source holds, on what evidence, and what remains genuinely unresolved. Reasons in depth on a bounded question. Does not split the difference, manufacture a reconciliation, or pick

  • source-read

    Use when a dispatch names one source — an installed package, a specification, a vendor document — and the question to put to it. Returns an answer grounded in quoted passages plus what the source does not say. Reads a source end to end. Does not generalize past the source or substitute recollection

Hooks (2)

  • PostToolUse

    "${CLAUDE_PLUGIN_ROOT}"/hooks/scripts/inject-extension.sh post-tool-use

  • UserPromptSubmit

    "${CLAUDE_PLUGIN_ROOT}"/hooks/scripts/inject-extension.sh user-prompt

View source on GitHub

More from it-bens/ai-tools

All 19 plugins →

llm-author

claudecodex

Author LLM-targeted content (prompts, skills, agents, rules files, session handoffs, and session feedback) for Claude 5 (including Opus 5.5, Sonnet 5.5, and Haiku 5.5) and Claude 4, GPT-6 (including GPT-6.1 Sol and Terra) and GPT-5.6 (OpenAI), GLM 4.7 (Z.ai), and Gemini 3 (including Deep Research)

5 skills · hooks

11

software-writer

claudecodex

Four universal, opinionated skills for writing software: writing-code (API-consultation gate, dependency shape, orphaned-symbol sweep, footgun check, two-tier comment policy with a code-comment consistency gate), writing-tests (classicist behavior-first workflow with a dead-code stop that reports to the caller), writing-docs (single-owner surfaces, pointer-file discipline, anti-slop validation via the human-author:ai-slop-writing-fixer subagent), and sweeping-dead-code (a report-only sweep for unused code across a path, the repository, or a branch: a definition ledger, read-only runs of installed dead-code tools, reachability classification, and independent refutation subagents). Each skill is extendable at every step via Pre-Step-N / Post-Step-N positions and via recognized named configuration values; the plugin ships hooks that deliver a project's extension files (.claude/extensions/software-writer/) wrapped in a structural envelope.

4 skills · hooks

11

Consult OpenAI Codex for fresh analytical perspective — auto-escalation when stuck, on-demand second opinions, and web research. Uses the official codex-mcp-server with session management and GPT-5.2-codex.

1 skill · 1 command · 1 agent · 1 MCP server

11

Debug AI tooling behavior by analyzing output against source instructions and enabling honest self-diagnosis.

2 skills

11

Portable MCP read_file tool for reading local text files with offset and limit arguments.

1 MCP server · hooks

11

Enforces use of Claude Code native tools / native-build bfs & ugrep instead of bash equivalents (cat, find, grep, sed, echo). Adapts messages to the detected build: new mode (macOS/Linux native with bfs/ugrep), classic mode (Windows, or older builds with Glob/Grep tools), or pass-through when neither toolchain is available.

1 skill · hooks

11

Similar plugins

superpowers

claudecodex

Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques

15 skills · hooks

obra/superpowers 316,836

claude-mem

claudecodex

Persistent memory for Claude Code: the next session starts where the last one left off.

29 skills · 1 MCP server · hooks

thedotmack/claude-mem 116,618

mem0

claudecodex

Cross-session memory and token savings for coding agents.

6 skills · 1 agent · 1 MCP server · hooks

mem0ai/mem0 68,893

context7

claudecodex

Up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your coding agent's context.

1 MCP server

upstash/context7 64,709

Build powerful, stateful CLI interfaces for any GUI application using the cli-anything harness methodology.

5 commands

HKUDS/CLI-Anything 58,665

Automates browser interactions for web testing, form filling, screenshots, and data extraction

12 skills

vercel-labs/agent-browser 50,373

Maintain work-orchestrator?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex plugins19 plugins