creating-skills
Use when you need to create a new custom skill for a profile - guides through gathering requirements, creating directory structure, writing SKILL.md, and optionally adding bundled scripts
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 4.9k
- Forks
- 590
01 前言 Front Matter
- name
- creating-skills
- description
- Use when you need to create a new custom skill for a profile - guides through gathering requirements, creating directory structure, writing SKILL.md, and optionally adding bundled scripts
03 SKILL.md 正文
<required> CRITICAL Add the following steps to your Todo list using TodoWrite:
- Gather skill requirements from me
- Select target profile
- Create skill directory structure
- Write SKILL.md with proper frontmatter
- (Optional) Write and bundle scripts
- Instruct the user to run /nori-switch-profile to switch profiles. </required>
Overview
This skill guides you through creating custom skills that persist across sessions. Skills are stored in profile directories and can include markdown instructions, checklists, and optional bundled scripts.
Writing Skills
Every skill must start with a required checklist block:
<required>
*CRITICAL* Add the following steps to your Todo list using TodoWrite:
1. <step 1>
2. <step 2>
...
</required>
This is the most important part of a skill.
Each step may have guidelines underneath. For example:
1. Create a directory.
Use `mkdir foo/bar`
2. Make a file.
...
Writing scripts
Skills may be bundled with scripts. Scripts are simple code cli tools that do various things deterministically.
Any scripts you write should be entirely self contained. Ask the user which language they prefer.
The scripts should be callable from the Bash tool.
The script should be stored in the same place as the skill. Add a section to the SKILL.md on how to use the script. If the script is required to be called, add that instruction to the <required> block.
Tools and Code
Document which tools are necessary for the skill.
The SKILL should explicitly encourage writing code. The agent should be told to write code to call a tool any time the agent needs to call any tools more than once. The agent should write code instead of calling tools itself.
Steps that reference tools and APIs should inline them in the step by step process.
<good-example>
- Write a python script that uses the slack api to pull my unread messages. The script should write my messages to a folder, /home/foobar/daily-analysis/<DD-MM-YYYY>/slack
- Write a python script that uses the gmail api to pull my inbox. The script should write my emails to a folder, /home/foobar/daily-analysis/<DD-MM-YYYY>/email
- Write a python script to identify the most urgent messages that I need to respond to. <system-reminder> Auth credentials can be found at /home/foobar/authentication/auth.txt </good-example>
<bad-example> Tools: Slack, Gmail API
- Summarize my messages by looking through common surfaces that I may use for messages. </bad-example>
<good-example>
Read .agency/plugins/nori/skills/finishing-a-development-branch/SKILL.md
</good-example>
<good-example>
Read .agency/plugins/nori/skills/webapp-testing/SKILL.md
</good-example>
04 同仓库更多 Skills
trigger-pipelines-for-copilot-pr
Trigger ADO pipelines for a Copilot-created PR by posting /azp run comments. Use when the user asks to trigger CI pipelines for a specific PR.
review
Review changes relative to a prior state, such as a diff, pull request, branch, commit, or working-tree changes. Use for change-focused code review and regression analysis. Do not use to assess an existing file, API, design, implementation, or other artifact without a comparison; if the request is ambiguous, ask whether the user wants a change review and what prior state to compare against.
review-pr-local
The authority on pulling a pull request — Azure DevOps or GitHub — into an isolated local worktree so a human can review it in their editor, and on posting human-controlled review comments back to the PR (ADO via the ado PR-thread MCP tools or the Azure DevOps REST API, GitHub via the gh CLI). Auto-detects the platform and all identifiers from the PR URL; not tied to any specific repository. The AI assists an interactive human review and only ever posts comments the human explicitly requests. If you find yourself checking out a PR branch over the user's own working tree, or posting AI-authored review prose the human did not ask for, invoke this skill instead.
policy-check
This skill should be used when the user asks to "run policy check", "check policy", "policy-check", or needs to validate package compliance. Provides guidance on running policy checks for specific packages or the entire repository.
pipeline-test-triage
Triage test failures in an Azure DevOps pipeline run for Fluid Framework. Use when the user shares an ADO build/pipeline URL or build ID and wants failures analyzed, when they ask "why did this pipeline fail", "summarize the test failures", "is this test flaky or a real bug", "compare these runs", or wants bugs filed for genuine failures. Covers the Real Service End to End Tests pipeline as the worked example but applies to any FF test pipeline.
fluid-release
Fluid Framework client release group — minor releases, patch releases, and post-release type test updates. Covers release prep, branching, version bumps, changelogs, release notes, and type test baselines. In autonomous mode, auto-detects state from the repo, attempts to execute, and falls back to a GitHub issue on failure. Triggers on "release", "do the release", "release status", version bump, release notes, changelog, release branch, or release engineering.