Skip to main content
skill CX Codex

intent · yeachan-heo — Claude Code skill: what it does, how to install

Turns a pasted chat or problem report into a reviewed intent.md. It gives /launch a decision brief instead of code.

MIT 187 installs ★ 40,997 updated 2026-10-24

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add Yeachan-Heo/oh-my-claudecode --skill intent -a claude-code
Open on GitHub

What intent does

The skill turns raw support or operations feedback into docs/intents/<slug>/intent.md. It covers five sections: the observable problem, goal, users and systems, constraints, and open questions. The agent shows missing details, asks numbered batches of questions, and checks each section before drafting.

The draft moves through four tracker records for submission, fact verification, the verdict, and execution. An accepted intent becomes the mission brief for /launch, while open questions carry into the next specification. The agent never signs or rejects on a person’s behalf, so human confirmation and review remain part of the workflow.

What it helps with

  • turn a support chat into a product problem statement
  • collect missing requirements in numbered question batches
  • draft docs/intents/<slug>/intent.md
  • route an intent through tracker review
  • hand an accepted intent to /launch

Example prompts for the agent

  • Turn this pasted chat into an intent and ask the missing questions in numbered batches.
  • Continue the open review from docs/intents/<slug>/intent.md.
  • Review this problem report, build the five sections, and prepare it for /launch.

How to install intent

1Skills CLI (Vercel)

npx skills add Yeachan-Heo/oh-my-claudecode --skill intent -a claude-code

2Manually: put the files in place

~/.claude/skills/intent/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/intent/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about intent

How do I install intent in Claude Code?

Run npx skills add Yeachan-Heo/oh-my-claudecode --skill intent -a claude-code, then check ~/.claude/skills/intent/SKILL.md.

What does the intent skill do?

It turns feedback or a chat log into a five-section intent.md, checks whether the sections are ready, and passes an accepted document to /launch as its mission brief.

Which platforms support intent?

The skill supports Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and OpenCode.

Source text

SKILL.md · MIT
---
name: intent
description: Shipyard's internal requirements intake for non-engineer contributors (support/ops) — turn a pasted chat log or verbal problem report into a five-section intent.md through numbered batch questioning with a completion gate, walk it through tracker review with harbor's four records, and hand the accepted intent to /launch as its mission brief. Produces decisions, not code.
argument-hint: "<pasted feedback / chat log | path to existing intent.md | nothing to continue open review>"
level: 3
disable-model-invocation: true
---

# Intent

Intent is the shipyard's **internal intake**: support/ops staff see product problems first but cannot write engineering specs. The intent skill converts their conversation into a goal-level YAML+markdown artifact that an owner reviews and signs, so engineering receives a fog-free mission. It stands on the verifiability boundary — the agent asks, records, and drafts; humans submit, verify, and sign.

Design provenance: every decision here was pinned by a wayfinder map (7 tickets) and walked end-to-end in a demo repository before implementation.

## Role contract

- The contributor (e.g. support staff) pastes raw feedback or a chat log. The agent never invents facts; unknowns become explicit `待确认` items.
- Decisions are the humans': the contributor confirms wording, the supervisor submits, the product owner verifies facts and signs accept/reject. The agent never signs, never rejects on anyone's behalf, and never stands in for a reviewer.
- The stop gate is a **completion gate, not an ambiguity score** — no math is borrowed from deep-interview; only its Goal and Constraint clarity judgments are reused as quality bars.

## The conversation (numbered batch questioning)

1. **Opening inventory**: restate the problem in one sentence, then show the five-section inventory (known / missing per section).
2. **Numbered batch questioning**: list every gap as numbered questions in one batch. The contributor answers by number (`1 是; 2 不确定`). Unknown answers are recorded as `待确认` and pushed into open questions — never pressed until guessed.
3. **Re-inventory**: apply the answers; if gaps remain, emit the next numbered batch. One round = one batch.
4. **Stop gate** (all five sections must pass):
   - 问题: observable phenomena, no solution embedded
   - 目标: one sentence, no qualifiers, names entities and relations (deep-interview Goal Clarity bar)
   - 用户和系统: real names for users and systems; unknowns written as `待确认` and counted as open questions
   - 约束: boundaries, limits, and non-goals explicit (deep-interview Constraint Clarity bar)
   - 未决问题: list non-empty, or an explicit `无`
5. **Soft limits**: soft prompt at round 6 ("draft with current clarity?"), hard cap at round 12. An early exit is never blocked — thin sections surface as longer open questions, and review is the gate that rejects them.
6. **Draft**: the agent drafts `docs/intents/<slug>/intent.md` (template below) at `status: draft`; the contributor confirms wording before any submission.

## The intent.md template

```markdown
---
intent: <slug>
title: <one-line title>
author: <name (role)>
date: <YYYY-MM-DD>
status: draft | in-review | accepted | rejected
round: <tracker review round>
---

## 问题
<observable phenomena, no solution>

## 目标
one sentence, no qualifiers

## 用户和系统
<who the users are, which systems, real names; unknowns as 待确认>

## 约束
<boundaries, non-goals; cite Rules pillar files when a constraint comes from them>

## 未决问题
- [阻塞|非阻塞] <question>   ← agent suggests, product owner finalizes
```

## Grading open questions

The test for `阻塞`: **"can the spec still be approved without answering it?"** No → blocking; yes → non-blocking (tracked, may ride into development). The drafting agent suggests grades; the product owner finalizes at spec approval.

## Review handoff (harbor four records)

Review runs on the repo's tracker — **the tracker is the only record source; intent.md frontmatter mirrors status only**. One record file per Intent, rounds accumulate inside it. Required fields per record: signer, date, verdict, link to intent.md, round, rejection reason (mandatory on reject).

- **proposal** = supervisor submits the Intent (intent.md → `status: in-review`)
- **verification** = product owner checks facts against repo evidence before deciding; the agent may draft the fact-check table, but the signature belongs to the reviewer — no repo evidence to check? Then verification records why none was needed
- **decision** = accept or reject; reject requires a reason, which is the key input to resubmission
- **execution result** = spec generation completed for this Intent

Rejected → revise → resubmit in the same file under the next round number. History stays traceable.

## Handoff to launch

An accepted intent (`status: accepted`) is a valid mission brief. Launch's spec synthesis then follows the **four-step contract**:

1. read `docs/intents/<slug>/intent.md` (must be accepted, latest round)
2. combine with the existing codebase
3. follow the Rules pillar (`CLAUDE.md` + `docs/standards/` + `docs/business/`)
4. list every doubt and rule conflict — graded into the spec's pending-confirmation section

Intent open questions carry into the spec verbatim, keeping their pending status.

## Amendment

New information after acceptance (typically surfaced by the first spec) is an **amendment**: anyone may initiate, the submitter signs, and it always re-enters full review as a new round — no minor-change exemption. An intent change marks the generated spec stale; the spec is regenerated from the new intent and re-approved. No separate amendment record type: rounds carry it.

## Conflict sedimentation

Conflict conclusions in the spec's pending section follow precedent-based routing: case-local answers stay in the spec; conclusions that will bind later Intents sediment into the Rules pillar (`docs/business/` for business rules, `docs/standards/` for behavior rules); hard-to-reverse technical tradeoffs become ADRs. Never double-write; never defer archiving. The product owner decides whether a conclusion is precedent-setting; the tech lead advises the destination.

© Yeachan-Heo · MIT · Yeachan-Heo/oh-my-claudecode

Similar skills

SKILL.md skills catalog →

intent by other authors comes first.

intent ghaida intent is a Claude Code skill that sets project context and routes UX work to specialized skills. It activates for product design, UX reviews, and questions about ethical design. CX Codex +1 3,400installs android-intent-security Google LLC The android-intent-security Claude Code skill audits Android component security and Intent handling. It runs for checks of AndroidManifest.xml or code that processes incoming Intents. CX Codex +1 2,400installs intent-layer crafter-station intent-layer builds a folder-by-folder context tree — one root AGENTS.md plus child nodes — so agents read your codebase the way a senior engineer would. CX Codex +1 639installs intent-modeling haowjy intent-modeling — a skill that makes an AI agent work out what you meant before it acts on what you literally said. CX Codex +1 586installs intent-framed-agent pskoett The intent-framed-agent skill writes down what a coding session is supposed to achieve before edits start, then watches for work that wanders off. Handy on long refactors and feature builds. CX Codex +1 440installs intent-signal-aggregator OneWave-AI intent-signal-aggregator — a skill that gathers buying signals — job posts, funding rounds, leadership changes — and ranks accounts by how ready they are to buy. CX Codex +1 331installs Keyword Research — Google Suggest, Intent & Long-Tail axel-belfort Keyword Research — Google Suggest, Intent & Long-Tail is an MCP server for Google Suggest keyword research, intent classification, and long-tail variants for content, PPC, and competitor analysis. CX Codex +1 6,583installs

From the same set48

The whole set →
agent-doc-discipline agent-doc-discipline sets writing rules for agent-facing specs, skills, and CLAUDE.md files. It ensures a fresh agent session acts correctly on instructions alone without needing prior chat context. CX Codex +1 196installs diagram This skill picks a compact visual for code structure, from a call tree to a Mermaid diagram. Use it when prose makes relationships hard to follow. CX Codex +1 194installs ask-navigator ask-navigator — a skill that turns an unclear effort into a map of decision tickets on the repo's issue tracker. It works the frontier one ticket per session until the route clears, then hands decisions over without building anything. CX Codex +1 191installs loft loft resolves deadlocked design and logic questions by building disposable runnable prototypes that never merge into main. CX Codex +1 189installs pr pr — a Claude Code skill that drafts pull request descriptions from the session paper trail. It compiles verified test evidence, visual summaries, and rollback risks into a markdown body. CX Codex +1 161installs map This Claude Code skill routes a task to the matching skill in the collection. It maps work to a delivery stage or a specific workflow, but does not carry out the task itself. CX Codex +1 160installs refit refit is a Claude Code skill that reviews OMC runtime traces to find agent workflow friction and propose targeted fixes to checks, configs, and instructions. CX Codex +1 158installs session-flow session-flow — a skill for Claude Code rebuilds the session’s goal, pivots, outcomes, current stage, and open decisions from its history. It starts after a session switch or context compaction. CX Codex +1 118installs

Project sponsors

Companies that keep this analytics open