ワンクリックでManusで任意のスキルを実行

始める

work

スター7

フォーク3

更新日2026年4月23日 13:28

Execute a plan produced by `/plan` — one task at a time, with tests and an independent review after each task.

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

Manusで実行

ソース

Aidbox

Aidbox/aidbox-hl7v2-example

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

Manusで実行

Work

Execute a plan file at ai/tickets/YYYY-MM-DD-*.md produced by /plan. The file starts with # Plan: [Name] and contains ## Task N: [Name] sections with checkboxes.

Cycle

Open the plan file. Pick the first task with unchecked checkboxes (start at Task 1 if nothing is checked).
Work through the task's checkboxes one at a time. Mark each as checked in the plan file immediately after completion.
Run the commands listed under ## Validation in the plan. If any fail, fix the root cause — don't skip.
Spawn an independent review agent (prompt below). Address findings, then re-run the same validation commands.
Print the full review text to the user (include addressed and ignored issues) and stop. Wait for the user to say "next task" before starting the next one.

If the task is the last one and the user approves, move the plan file to ai/tickets/completed/.

Review agent spawn

Use the Agent tool (not the Skill tool — this is a sub-agent running ai-review, not you invoking it yourself). Substitute [N] and [plan-file-path]:

Agent({
  description: "Review Task [N]",
  subagent_type: "general-purpose",
  prompt: "Use skill ai-review to review implementation of Task [N] from [plan-file-path]. Think hard. The changes are uncommitted. Return your review output; do not change any files."
})

If the user specified codex for reviews, instead run via the Bash tool:

codex exec --model gpt-5.3-codex --sandbox workspace-write --full-auto "<prompt>"

Rules

One task per turn. Never start task N+1 in the same turn you finish N.
Don't plan ahead. The plan file is the plan. Don't maintain your own TODO list.
Never skip tests. "Validation passed" means you ran them and they passed. If you can't run them, report the blocker — don't mark the checkbox.
Fix failing tests, even pre-existing ones. If a test's expected behavior is wrong, ask the user before changing it.
Always spawn a review agent. Don't review your own code.

このリポジトリの他の Skills

同じリポジトリ

check-errors

Aidbox/aidbox-hl7v2-example

Check recent HL7v2 processing errors in Aidbox, diagnose root causes, suggest fixes. Use when asked to check errors, diagnose failures, or troubleshoot message processing.

2026-04-247

fhir-info

Aidbox/aidbox-hl7v2-example

Look up FHIR R4 resource/datatype fields, types, cardinality, reference targets, and coded enums (including custom resources like IncomingHL7v2Message, OutgoingBarMessage). Use when writing converters or verifying FHIR compliance.

2026-04-247

hl7v2-to-fhir-pipeline

Aidbox/aidbox-hl7v2-example

Build a new HL7v2→FHIR message converter or extend an existing one. Orchestrates requirements gathering, then delegates to `/plan` and `/work`.

2026-04-247

plan

Aidbox/aidbox-hl7v2-example

Create a checkbox task plan for a feature or fix. Output is a markdown file that `/work` executes one task at a time.

2026-04-237

aidbox-request

Aidbox/aidbox-hl7v2-example

Authenticate ad-hoc curl requests to the local Aidbox using the root Client secret from docker-compose.yaml. Use for debugging and one-off queries from the shell (not from app code — `src/aidbox.ts` already wraps auth).

2026-04-227

hl7v2-info

Aidbox/aidbox-hl7v2-example

Look up HL7v2 segment fields, datatypes, components, tables, and message structures. Use when writing converters, checking field positions/optionality, or verifying HL7v2 spec compliance.

2026-04-227

Work

Execute a plan file at ai/tickets/YYYY-MM-DD-*.md produced by /plan. The file starts with # Plan: [Name] and contains ## Task N: [Name] sections with checkboxes.

Cycle

Open the plan file. Pick the first task with unchecked checkboxes (start at Task 1 if nothing is checked).

Work through the task's checkboxes one at a time. Mark each as checked in the plan file immediately after completion.

Run the commands listed under ## Validation in the plan. If any fail, fix the root cause — don't skip.

Spawn an independent review agent (prompt below). Address findings, then re-run the same validation commands.

Print the full review text to the user (include addressed and ignored issues) and stop. Wait for the user to say "next task" before starting the next one.

If the task is the last one and the user approves, move the plan file to ai/tickets/completed/.

Review agent spawn

Use the Agent tool (not the Skill tool — this is a sub-agent running ai-review, not you invoking it yourself). Substitute [N] and [plan-file-path]:

Agent({ description: "Review Task [N]", subagent_type: "general-purpose", prompt: "Use skill ai-review to review implementation of Task [N] from [plan-file-path]. Think hard. The changes are uncommitted. Return your review output; do not change any files." })

If the user specified codex for reviews, instead run via the Bash tool:

codex exec --model gpt-5.3-codex --sandbox workspace-write --full-auto "<prompt>"

Rules

One task per turn. Never start task N+1 in the same turn you finish N.

Don't plan ahead. The plan file is the plan. Don't maintain your own TODO list.

Never skip tests. "Validation passed" means you ran them and they passed. If you can't run them, report the blocker — don't mark the checkbox.

Fix failing tests, even pre-existing ones. If a test's expected behavior is wrong, ask the user before changing it.

Always spawn a review agent. Don't review your own code.

name	work
description	Execute a plan produced by `/plan` — one task at a time, with tests and an independent review after each task.