Skip to content
Agentmart
Submit a tool

dispatch

Multi-agent planning, review, implementation.

Skill by Gyunikuchan

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/gyunikuchan-dispatch

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Multi-agent planning, review, implementation. Use only when explicitly invoked.

Use cases: agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found
Runs as
Node.js package

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/dispatch

What's inside

Skill name
dispatch
Description
Multi-agent planning, review, implementation. Use only when explicitly invoked.
License
MIT
Instructions
71 lines in SKILL.md · view on GitHub

Before installing

  • Ships executable scripts

Files (139)

  • README.md
  • SKILL.md
  • config.sample.jsonc
  • change-handling.md · references/
  • decision-frame.md · references/
  • diagnostics.md · references/
  • glossary.md · references/
  • providers.md · references/
  • ask.md · references/readme/
  • concepts.md · references/readme/
  • configuration.md · references/readme/
  • design.md · references/readme/
  • implement.md · references/readme/
  • plan.md · references/readme/
  • review.md · references/readme/
  • troubleshooting.md · references/readme/
  • review-rules.md · references/
  • self-build.md · references/
  • spec-stage.md · references/
  • design.md · references/templates/
  • plan.md · references/templates/
  • quality-bar.md · references/templates/
  • review-prompt-code.md · references/templates/
  • review-prompt-design.md · references/templates/
  • review-prompt-plan.md · references/templates/
  • review-prompt.md · references/templates/
  • report-code.json · references/templates/schemas/
  • report-design.json · references/templates/schemas/
  • report-plan.json · references/templates/schemas/
  • spec.md · references/templates/
  • walkthrough.md · references/templates/
  • write-brief-hotfix.md · references/templates/
  • write-brief-task.md · references/templates/
  • write-brief.md · references/templates/
  • ask.md · references/verbs/
  • design.md · references/verbs/
  • implement.md · references/verbs/
  • plan.md · references/verbs/
  • review.md · references/verbs/
  • worktree-cleanup.md · references/
  • checkpoint.ts · scripts/core/
  • diagnostics.ts · scripts/core/
  • effect-id.ts · scripts/core/
  • effect-recovery.ts · scripts/core/
  • frame.ts · scripts/core/
  • interpreter.ts · scripts/core/
  • journal.ts · scripts/core/
  • lock.ts · scripts/core/
  • ports.ts · scripts/core/
  • progress.ts · scripts/core/
  • spec-discovery.ts · scripts/core/
  • types.ts · scripts/core/
  • validate.ts · scripts/core/
  • dispatch.ts · scripts/
  • criterion-id.ts · scripts/domain/
  • design.ts · scripts/domain/
  • diagnostics.ts · scripts/domain/
  • effect-digest.ts · scripts/domain/
  • engine-fault.ts · scripts/domain/
  • execution-config.ts · scripts/domain/
  • …and 79 more

SKILL.md (first lines)

# Dispatch

## Invocation

Parse `[level] [(pins)] [verb:] argument`; default `ask`. Verbs are `ask`, `design`, `plan`, `implement`, `review`. Use design for ordered increments, plan for one unit, implement for delivery. An empty code-review argument means `worktree`; `.plan.md` and `.design.md` infer review kind.

Pick and report the level. Preserve explicit levels; `xhigh` and `max` require user selection. Levels select roster, rounds and approval need. Use configured models and efforts exactly; provider configuration owns roster selection.

Initialize with `node <skills-dir>/dispatch/scripts/dispatch.ts session init --objective "<objective>"`. Retain the returned session path and follow any returned `instructions`. Reuse with `session reactivate --session-dir <session>`; `session quota-reprobe` refreshes one quota key. See [CLI contract](references/toolkit.md) for session options and outputs.

## Tool loop

Invoke tools with `node <skills-dir>/dispatch/scripts/dispatch.ts <tool> <action> … --session <session>`. Stdout contains `ok`, a summary of at most 1024 UTF-8 bytes, and a result `path`. Read that file for details. Exit codes: 0 operation returned (inspect `ok`), 1 usage or invalid input, 2 fault, 3 lock held, 4 gate or pending native work.

1. Read host rules and the [quality bar](references/templates/quality-bar.md). Resolve the [governing spec](references/spec-stage.md).
2. Author and lint the required design or plan; resolve parse errors. Review against the spec, adjudicate findings and apply accepted fixes under [review rules](references/review-rules.md).
3. Obtain the applicable approval tuple under the [Decision Frame](references/decision-frame.md).
4. Implement scoped tasks with native writers, then verify the integrated tree. Inspect failed checks and repair within scope. Resolve code findings, commit fixes, verify and review the exact HEAD. For transient plans, run `worktree prune`, then `verify run` with final selection under the [lifecycle](references/verbs/implement.md). Deliver only after successful final checks.
5. Run `report complete`; after success run `worktree cleanup` per [cleanup](references/worktree-cleanup.md). Report behavior, evidence, concerns and artifact paths.

For `native-pending`, read each `promptPath`, launch the configured native slot, write its output to `replyPath`, and rerun the same tool call. Preserve running handles; wait on completion or material blockers. [Providers](references/providers.md) owns slot and fallback rules.

## Invariants and recovery

User-only records (approval, waiver, manual completion and selection of xhigh/max) require actual user quotes. Never invent decisions. At medium and above, writer briefs and integration require the approved current artifact tuple or its recorded non-intent edit chain. Intent, external dependency/runtime prerequisite and discarded-work changes need new approval.

Provider delegates stay read-only. Native writers edit only assigned paths and run no Git write commands. Preserve unrelated dirty and ignored files. Tools own ledger and result files; append records through the CLI.

After interruption, use `status show`, inspect its result and retained worktrees, then resume the next unfinished operation. Keep uncertain effects and handles until completion is established.

Read the phase you execute: [ask](references/verbs/ask.md), [design](references/verbs/design.md), [plan](references/verbs/plan.md), [implement](references/verbs/implement.md), [review](references/verbs/review.md). For changes to Dispatch itself, use [self-build](references/self-build.md).

From Gyunikuchan/dispatch-skills, MIT.

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Oct 21, 2026 (7 days ago)
Last release
Oct 21, 2026 (7 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from Gyunikuchan

Cite this page

dispatch skill. Agent Mart. https://agentmart.dev/skills/gyunikuchan-dispatch (last updated 2026-10-24)