Skip to content
Agentmart
Submit a tool

design-code-implement

Turn a settled requirement into at least three grounded implementation directions, then record the chosen one.

Skill by GUOJIE-HONG

IndexedMIT★ 1
Write a review

Chalk Labs

Need help deploying design-code-implement?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/guojie-hong-design-code-implement

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Turn a settled requirement into at least three grounded implementation directions, then record the chosen one. Use after the requirement is settled and before any code is written.

Use cases: design

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/design-code-implement

What's inside

Skill name
design-code-implement
Description
Turn a settled requirement into at least three grounded implementation directions, then record the chosen one. Use after the requirement is settled and before any code is written.
License
MIT
Metadata
internal: true
Instructions
90 lines in SKILL.md · view on GitHub

Files (3)

  • SKILL.md
  • openai.yaml · agents/
  • template.md · references/

SKILL.md (first lines)

# Design Code Implement

Decide *how* to build something whose *what* is already settled. Explore the repository's real architectural conventions with parallel sub-agents, surface where the new requirement is in tension with them, and offer at least three directions the user picks from.

This skill does not write production code. It ends when a direction is chosen and recorded.

## 1. Establish the input

Accept the requirement in whatever form it arrives: a file path, an upstream skill's handoff, or prose in the conversation. Never demand a file. Read what the user points to, and restate the requested outcome in one sentence; get it confirmed if it is at all ambiguous.

Do not ask the user for facts discoverable from the repository, runtime, or tools.

## 2. Scope the exploration

Decide the aspects yourself, from the requirement confirmed in §1. Do not send a scout sub-agent first, and do not ask the user to approve the aspect list — they cannot judge it before seeing the repository's reality, and a wrong aspect exposes itself in the reports.

Constraints on the aspect list:

- **Always include domain language**: what the existing concepts are called, and whether the requirement introduces a term the codebase does not have.
- **Architectural convention only**: how layers are cut, how errors travel upward, how data enters and leaves, where tests live, where the existing seams are. Exclude surface style — naming case, file placement, formatting. Linters and neighbouring code already teach those, and nobody picks a different implementation direction because of camelCase.
- **At most five sub-agents.** If the requirement seems to need more, merge aspects. A requirement genuinely spanning eight architectural aspects should be split before it is designed.
- **Bounded blast radius.** Explore only the areas the requirement touches, plus recent hotspots from `git log --oneline`. Do not sweep the whole repository; irrelevant conventions dilute the real tensions.

## 3. Dispatch and collect

Give every sub-agent the same tool preference:

- If a codebase knowledge-graph MCP is available, use it first; complete its indexing if the repository is not indexed yet.
- Otherwise prefer any other repository-aware navigation tool.
- Fall back to built-in search for string literals, configuration, scripts, and non-code files, or when the above return stale or inapplicable results.

Require every sub-agent to report exactly three things per finding:

1. **The existing convention** — what this repository already does for this aspect.
2. **The evidence** — file and line (`path/to/file.ts:42`). A claim without a path is not usable.
3. **The tension** — where the new requirement collides with that convention. If there is none, say so explicitly.

The tension field is the point of the exercise. Conventions with no tension are context; tensions are where directions come from.

## 4. Produce the directions

**If exploration found no existing convention for the requirement to inherit**, say so and stop; this skill has nothing to ground directions in. Do not produce directions.

Present **at least three** directions in the conversation, sequentially, so each is absorbed before the comparison.

- **Direction 1 is always the baseline**: follow every existing convention, open no new seam, take the most conservative path. It costs almost nothing to produce and it is what the other directions are measured against — without it the user cannot judge whether the extra cost is worth paying.
- **Directions 2 and 3 grow out of the reported tensions**, not from a template. Do not assign directions predetermined constraints; let the repository's actual friction decide where they diverge.

Give each direction exactly four fields:

1. **Positioning** — one sentence.
2. **Footprint** — which files and which seams it touches.
3. **Cost** — what it charges, concretely.
4. **When this is the wrong choice** — the condition under which this direction is a mistake.

The fourth field is mandatory and forces honesty. A direction with no failure condition has not been thought through.

Close with a **recommendation**: the direction you would pick and the tension or requirement that decides it. The choice stays the user's.

**When the tensions cannot support three real directions**, say so plainly: report that the existing conventions already determine the approach, and stop. Never pad the list with contrived variants — a user who cannot tell a real choice from a manufactured one stops trusting all of them.

**Mixing is allowed.** If the user wants one direction's overall shape with another's error handling, take it — then restate the combined direction in full and get it confirmed before writing anything. A mix understood differently by each side is worse than no mix.

## 5. Record the decision

Write `design.md`.

**Placement**: ask the user where it goes before writing it.

**Contents** — only what will be done, laid out as in [`references/template.md`](references/template.md):

- The chosen direction as diagrams, not prose.
  - Each flow that crosses components: an ASCII flowchart, one per trigger. Steps run top to bottom, numbered, each followed by an indented `↳` line naming the real function and file that does it. Leave the right side open, with only left-edge arrows and indentation, so wide characters never break alignment. Draw the success path only; under the diagram, list every failure the requirement implies as "step N fails → what happens". A step that is a flow of its own gets its own diagram, referenced as "→ Flow N".
  - Each entity that changes state: a Mermaid `stateDiagram-v2`.
  - Changes no diagram shows (migrations, configuration, interface shapes): a short list.
- The conventions this work must follow, one line each with its evidence path.
- Nothing else. **Do not write the rejected directions, and do not write their rationale.** A downstream implementing agent reads this file as instructions; describing an approach that is not being taken invites it to be taken.

**If `design.md` already exists**: read it first, then update it — carry forward whatever still holds. Never overwrite it unseen.

## 6. Hand back and stop

Report:

- The chosen direction in a short summary.
- The path to `design.md`.

Then stop. How and with what to implement is the user's call; do not start implementing or name a next command.

From GUOJIE-HONG/skills, MIT.

More from GUOJIE-HONG/skills

Signals

Stars
1
Forks
1
Open issues
1
Last commit (this folder)
Oct 10, 2026 (20 days ago)
Last release
Oct 23, 2026 (8 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from GUOJIE-HONG

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get design-code-implement running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying design-code-implement?

CallGet help
Cite this page

design-code-implement skill. Agent Mart. https://agentmart.dev/skills/guojie-hong-design-code-implement (last updated 2026-10-24)