Guide · Agent skill
How to use the Intent Intake agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Intent skill is an agent skill built for Shipyard's internal intake, where support and operations staff describe product problems that engineers then need to act on. It turns a pasted chat log or a spoken problem report into a five-section intent.md document, asking numbered batches of questions along the way. After this guide you will know what the skill does, how to install it, how to start a session, and how the document moves through review.
Intent IntakeYeachan-Heo/oh-my-claudecode · MIT Download the skill Quick start
- 1Download the ZIP. Get the Intent skill ZIP from the directory page.
- 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for personal use or into your project's .claude/skills/ folder. Alternatively, upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
- 3Confirm the repo layout. Make sure your repository has a docs/intents/ folder and a tracker for review records.
- 4Start the session. Invoke the skill with pasted feedback, a path to an existing intent.md, or nothing to continue an open review.
- 5Answer numbered batches. Reply by number to each batch. Use 不确定 for anything unknown so it is recorded as an open question instead of guessed.
- 6Submit and review. Confirm the drafted wording, then submit so the intent moves through tracker review and, once accepted, into /launch.
What the Intent skill makes the agent do
An agent skill is a folder of instructions (a SKILL.md file) that tells Claude how to handle a specific kind of task. The Intent skill covers requirements intake: it converts raw feedback into a structured intent.md file. The agent restates the problem, shows what is known and missing in each of five sections, and asks every gap as a numbered question in one batch.
The agent does not guess. Answers you cannot give are recorded as 待确认 (pending confirmation) and listed as open questions. The output is a decision document, not code.
- Five sections: 问题 (problem), 目标 (goal), 用户和系统 (users and systems), 约束 (constraints), 未决问题 (open questions)
- Questions are asked in numbered batches; you answer by number, for example 1 是; 2 不确定
- Output is drafted at status: draft at docs/intents/<slug>/intent.md
Install the Intent skill
Download the skill ZIP from the directory page. You can install it in one of two ways. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.
In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Check what you need before you start
The skill is written for Shipyard's internal process and assumes a repository with a docs/intents/ folder and a tracker for review records. The workflow also refers to the /launch skill and the harbor review process, which are not included in the skill file. If your constraints cite project standards, the skill expects pillar files such as CLAUDE.md, docs/standards/ and docs/business/ to exist.
- A repository with a docs/intents/ folder
- A tracker for review records
- The /launch skill and harbor review process, if you want the full handoff
- Pillar files (CLAUDE.md, docs/standards/, docs/business/) if constraints reference them
Start an Intent session and answer the batch questions
The skill file sets disable-model-invocation: true, so invoke it explicitly rather than relying on the agent to pick it up on its own. The argument hint accepts three inputs: pasted feedback or a chat log, a path to an existing intent.md, or nothing when you want to continue an open review.
After the opening inventory, reply to the numbered batch by number. Each round is one batch. The agent offers a soft prompt at round 6 asking whether to draft with current clarity, and has a hard cap at round 12. You can exit early at any time; thin sections become longer open questions that reviewers can reject.
- Paste the chat log or describe the problem in one or two sentences to start
- Answer by number; write 不确定 for anything you do not know
- Expect the stop gate to check all five sections before drafting
1 是; 2 不确定Review, sign-off and handoff to /launch
Once you confirm the wording, the supervisor submits the Intent, which moves intent.md to status: in-review. Review runs on the repository's tracker, which is the only record source. Each Intent has one record file with four record types: proposal, verification, decision and execution result. Rejections need a reason, and resubmission happens in the same file under the next round number.
An accepted intent (status: accepted) can be passed to /launch as its mission brief. Open questions carry into the generated spec with their pending status intact.
- proposal: supervisor submits the Intent
- verification: product owner checks facts against repo evidence
- decision: accept or reject, with a reason on reject
- execution result: spec generation completed
Safety notes: scripts, network and sign-off
The skill runs no scripts and uses no network access, according to its listing. It works by reading and writing files in your repository. The skill is designed so the agent never signs, accepts or rejects on anyone's behalf. Verification and decisions belong to named human reviewers.
Because the content reflects Shipyard's internal workflow, review the drafted intent.md before submitting it, and check that the open questions are graded correctly.
- No scripts are run
- No network use
- The agent drafts and asks; humans verify, submit and approve
- Blocking questions are graded by asking whether the spec can still be approved without answering them
What people use it for
- Turning a support chat log about a checkout error into a five-section intent.md for engineering
- Capturing a verbal operations complaint as a reviewable decision document with open questions listed
- Revising a rejected intent by resubmitting it under the next review round
- Using the five-section structure as a template for another team's requirements intake
Frequently asked questions
Is the Intent skill free to use?
The skill is published under the MIT licence, as stated on its listing. The listing does not mention pricing, so check the directory page for any other terms.
What do I need to run the Intent skill?
You need a repository with a docs/intents/ folder and a tracker for review records. The workflow also refers to the /launch skill and the harbor review process, which are not included in the skill file.
Where do I install the Intent skill?
Unzip the downloaded ZIP into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
Does the Intent skill run scripts or access the network?
According to the listing, it runs no scripts and uses no network. It reads and writes files in your repository.
Can the agent approve my intent automatically?
No. The skill is designed so the agent never signs or accepts on anyone's behalf. The product owner verifies facts and makes the accept or reject decision.
Is this skill useful outside Shipyard?
The content is written for Shipyard's internal process, so outside teams would need to adapt the structure and workflow to their own tracker and repository layout. It is not a general-purpose spec writer, and it produces decisions rather than code.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
The Intent skill suits support or operations staff at Shipyard who report product problems and need a structured, reviewable handoff to engineers. Teams wanting a template for their own requirements intake can adapt it, but it will not work out of the box without Shipyard's repository layout, tracker, and /launch and harbor workflows. If you need code written directly or automatic approval, pick a different tool.
Download the skill