Skip to content

SepineTam/mcp-for-stata / .claude/skills/rfc-impl-generator

rfc-impl-generator

Verified

Generate RFC and IMPL documents from a user-provided feature/fix description. Use when the user says something like "I want to add a feature with codex", "Implement this", "Write an RFC for...", or provides a short idea that needs to be formalized into the ai-driven/rfc + ai-driven/impl pipeline.

---
name: rfc-impl-generator
description: Generate RFC and IMPL documents from a user-provided feature/fix description. Use when the user says something like "I want to add a feature with codex", "Implement this", "Write an RFC for...", or provides a short idea that needs to be formalized into the ai-driven/rfc + ai-driven/impl pipeline.
allowed-tools: Read Grep Write Edit AskUserQuestion
---

# RFC/IMPL Document Generator

Convert a user's feature or fix idea into a paired RFC (design document for humans) and IMPL (implementation prompt for AI) in the `ai-driven/` directory.

## When to Use

- User says "I want to add X feature"
- User describes a bug fix or improvement
- User says "write an RFC for..."
- User gives a short idea that needs formalization before coding

## Workflow

### Step 1. Receive the User's Description

The user provides a short description of what they want. This could be a sentence, a paragraph, or a reference to a GitHub issue.

### Step 2. Assess Information Completeness

Determine if the description is sufficient to write a meaningful RFC and IMPL. Ask yourself:

- Do I understand **what problem** this solves?
- Do I understand **what the user expects to happen** after the change?
- Do I understand **when/where** the user would use this?
- Do I know the **scope** (what's in, what's out)?

If any of these are unclear, proceed to Step 3. If all are clear, jump to Step 4.

### Step 3. Ask Clarifying Questions (if needed)

Use `AskUserQuestion` to gather missing information. **Focus on expected outcomes and user experience, NOT technical implementation details.**

**Good questions (ask these):**
- "What should the user see or experience after this change?"
- "When would someone use this? Can you walk through a concrete scenario?"
- "What problem does this solve for you?"
- "How important is this compared to other features you're considering?"

**Bad questions (avoid these):**
- "Should this be a new parameter or a config option?"
- "Do you want to use threading or multiprocessing?"
- "What data structure should we use?"
- "Should we cache this in Redis or a file?"

**Rules for asking:**
- Ask 1-2 questions at a time
- After each answer, re-assess if you have enough information
- If still unclear, ask follow-up questions
- Stop once you can write a coherent RFC and IMPL

### Step 4. Generate the Documents

Once you have sufficient information, generate both documents.

#### 4.1 Determine the filename

Use the naming convention from `.claude/rules/ai-driven-coding.md`:
- If there's a GitHub issue: `{issue-number}-{short-name}.md`
- If no issue: `{prefix}-{short-name}.md` (prefix = feat-, fix-, refactor-, perf-, docs-)

`short-name` should be 2-4 English words, kebab-case.

**Before writing, ask the user to confirm the filename** if you're unsure about the issue number or prefix.

#### 4.2 Write the RFC

Create `ai-driven/rfc/{filename}`. Read [references/rfc-template.md](references/rfc-template.md) for the exact structure and copy it as the starting scaffold.

**RFC rules:**
- Written in the user's natural language
- No concrete code (no diffs, no real implementation)
- No "future work" sections
- Focus on "what" and "why", not "how"

#### 4.3 Write the IMPL

Create `ai-driven/impl/{filename}`. Read [references/impl-template.md](references/impl-template.md) for the exact structure and copy it as the starting scaffold.

**IMPL rules:**
- Written in English (AI-facing)
- Pseudocode only, no real runnable code
- Use "suggested", "recommended", "consider" — leave room for AI judgment
- Must end with a Checking List

### Step 5. Present to User

After generating both files:
1. Show the user the two file paths
2. Summarize what was written in 2-3 sentences
3. Ask if they want to review or modify anything before handing off to a Coding Agent

## Reference Files

| Name | Location | Description |
|:---|:---|:---|
| RFC Template | `references/rfc-template.md` | Empty RFC scaffold to copy for new design docs |
| IMPL Template | `references/impl-template.md` | Empty IMPL scaffold to copy for new AI prompts |
| RFC Example | `references/rfc-example.md` | Filled RFC for issue #60 (exit-clear-unix) |
| IMPL Example | `references/impl-example.md` | Filled IMPL for issue #60 |
| Workflow Example | `references/workflow-example.md` | Full walkthrough from vague idea to RFC+IMPL pair |
---
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall