パックの内容

SKILL.md（手順書）1点とスターターテンプレート1点が入っています。プレビューはこちらです：

Open-Source Coding Agent Skills

When to use

• You want a coding agent whose behaviour you can inspect, configure, and change — not a black box whose settings live somewhere you cannot reach
• You need to choose between model providers, or run a local model, and want that choice made deliberately instead of by default
• Your agent keeps editing files it should not touch, running commands it should not run, or burning through usage on unfocused work
• You want the same agent setup on every machine you own, reproducible from a config file instead of rebuilt from memory
• You need an audit trail: exactly what the agent changed, why, and who approved it

Workflow

1. Pick the agent build and provider (input: your constraints → output: config decision)
   Decide on paper before you install anything: what the agent must be able to do (read, edit, run commands, browse), what it must never do (delete, push, touch secrets), whether it runs against a hosted model or a local one, and what your budget ceiling is. The decision is a short written list, not a vibe — it becomes the config in the next step, and it is what you argue against when a shiny new option tempts you mid-project.
2. Write the config once (input: config decision → output: config file)
   Translate the decision into the agent's config file: provider and model, allowed tools, working directory limits, command allowlist, and usage caps. Keep the config in the project repo so it is versioned like any other file. If a setting cannot be expressed in the config, it does not exist — a rule that lives only in a chat message will be forgotten by the next session.
3. Define local working rules (input: repo → output: rules file)
   Write the repo's working rules in a file the agent reads at session start: code style, test commands, commit conventions, files that are generated and must never be hand-edited, and the review-gate rule from step 4. Keep it short — a rules file nobody reads is worse than no rules file, because it creates the illusion of rules. Ten lines that are followed beat fifty that are not.
4. Run the review-gated loop (input: task → output: gated change)
   Every task runs the same loop: the agent proposes a plan, you approve the plan, the agent makes the change, the checks run, you review the diff. No step is skipped because the task is small — small tasks are where bad habits form. The gate is not bureaucracy; it is the point where a human who understands the project looks at what a fast machine produced.
5. Audit what the agent touched (input: session → output: audit note)
   After each session, list what changed: files, commands run, and anything the agent attempted that was blocked. Read the blocked attempts — they tell you where your config or rules are unclear. If the same block keeps firing, the rule is wrong, not the agent; fix the rule.
6. Keep the setup reproducible (input: working setup → output: setup doc)
   Write the setup doc: how to install the agent, where the config lives, which provider credentials are needed and where they are stored, and how to verify a fresh install works (one smoke-task command and its expected result). A setup that only works on one machine is a hobby; a setup doc makes it infrastructure.

テンプレートのプレビュー

Open-Source Coding Agent — Setup Template

One repo = one config, one rules file, one setup doc. Fill this in before the first agent run.

Config decision

Agent build: {which agent, which version}

Model provider: {hosted / local} | Model: {name} | Budget ceiling: {per day / per month}

Agent may: {read / edit / run commands / browse — list exactly}

Agent must never: {delete / push / touch secrets / other bans}

Config file

Location in repo: {path to config file}

Provider / model: {as configured}

Allowed tools: {list}

Working directory limit: {path}

Command allowlist: {commands the agent may run}

スキルパック本体（SKILL.md とテンプレート）は英語で書かれています。

スキルパックを購入 — $4.99 Pro — 全511件のスキル