Skip to content
Agentmart
Submit a tool

understand

Clarifies scope before code is written.

Skill by AQaddora

IndexedMIT★ 3
Write a review

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/aqaddora-understand

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

Clarifies scope before code is written. Summarizes existing context, identifies gaps, and asks targeted clarifying questions one at a time so the AI doesn't sprint in the wrong direction. Triggers on "understand this", "what am I missing", "clarify requirements", "make sure you got it", or any vague feature request. Produces a structured comprehension brief that the user approves before any code is generated, and offers to save useful clarification patterns to ~/ai-doctrine.md so every other AI tool (Cursor, Copilot, Codex, Gemini) starts with the same comprehension habit.

Use cases: agent-building

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/understand

What's inside

Skill name
understand
Description
Clarifies scope before code is written. Summarizes existing context, identifies gaps, and asks targeted clarifying questions one at a time so the AI doesn't sprint in the wrong direction. Triggers on "understand this", "what am I missing", "clarify requirements", "make sure you got it", or any vague feature request. Produces a structured comprehension brief that the user approves before any code is generated, and offers to save useful clarification patterns to ~/ai-doctrine.md so every other AI tool (Cursor, Copilot, Codex, Gemini) starts with the same comprehension habit.
License
MIT
Instructions
110 lines in SKILL.md · view on GitHub

Files (3)

  • SKILL.md
  • comprehension-brief.md · references/
  • README.md · scripts/

SKILL.md (first lines)

# Understand

The second letter of **PURE**. The phase that saves you from shipping the wrong thing fast.

Most AI failures aren't generation failures. They're comprehension failures — the AI built exactly what you asked for, and what you asked for wasn't what you needed. This skill closes that gap before any code is written.

## When to Use This Skill

Trigger when the user:

- Says "understand this", "what am I missing", "clarify this", "make sure you got it"
- Gives a vague feature request ("add user search")
- Hands over a long thread or file and asks "do you have the context?"
- Asks the AI to confirm scope before coding
- Says "ask me what you need" or "what's unclear?"

## How It Works (4-Step Protocol)

### Step 1 — Restate

Restate the task back to the user in your own words, in two layers:

1. **Surface restatement** — what the user literally asked
2. **Inferred intent** — what they probably mean given context (project, doctrine, conversation history)

Surface gaps where surface ≠ inferred.

### Step 2 — Catalogue ambiguities

List every assumption your restatement implicitly makes. Group them by:

- **Blocking** — would change the solution architecture (e.g., "is this a one-time script or production endpoint?")
- **Material** — would change the implementation details (e.g., "do we need pagination?")
- **Minor** — would change naming or comments only

### Step 3 — Ask one question

Present blocking ambiguities first, **one at a time**. Never fire a five-question survey. After the user answers, re-evaluate — sometimes one answer collapses three other questions.

### Step 4 — Produce the comprehension brief

When ambiguities are resolved, output a structured brief:

```markdown
## What I'll Build

**Goal:** [one sentence]
**Inputs:** [what comes in]
**Outputs:** [what goes out]
**Constraints:** [perf, security, compatibility]
**Out of scope:** [what we are NOT doing this round]
**Open questions:** [if any remain that aren't blocking]
```

Wait for the user to approve before handing off to Refine or Execute. Offer to save valuable clarification patterns to `~/ai-doctrine.md`.

## Files in This Skill

| File | Purpose |
| --- | --- |
| `SKILL.md` | This file |
| `references/comprehension-brief.md` | Canonical brief template — TODO: annotated examples |
| `scripts/README.md` | Scripts placeholder |

## Operational Rules

1. **Restate before you ask.** Reading back what you heard catches half the ambiguities for free.
2. **One question at a time.** Five questions = the user picks the easy one and skips the hard one.
3. **Block on blockers.** Don't move to Refine/Execute until blocking ambiguities are resolved.
4. **Surface assumptions explicitly.** "I'm assuming X" is a real statement; "I'll do X" without flagging the assumption is a future bug.
5. **Time-box.** If clarification takes >5 messages, the task is too big — split it.

## Doctrine Integration

This skill captures two kinds of doctrine-worthy patterns:

1. **Clarification prompts that worked** — Save as **Understand**-category rules. ("Always ask 'is this a one-time script or production?' for any data-processing request.")
2. **Common ambiguity classes** — Save as **Understand** rules. ("In this codebase, 'user' is ambiguous — clarify customer vs admin vs internal_user.")

Save flow routes through `doctrine-keeper`. After the next `sync_to_agents.sh`, the same comprehension habit applies in Cursor, Copilot, Codex, Gemini, and Windsurf.

## Example Invocation

**User:** "Add user search to the dashboard."

**Skill activates:**

> Surface: "Add a user-search feature to the dashboard."
> Inferred (from your CLAUDE.md): admin-facing dashboard, server-side search, must filter by tenant.
>
> Before I plan: is this **substring search on name + email**, or **fuzzy search across all profile fields including custom attributes**? The two need very different indexing strategies.

*[User clarifies.]*

> Got it. Here's the brief — approve and I'll move to Refine:
>
> **Goal:** Add a substring search on user.name + user.email filtered by current admin's tenant.
> **Inputs:** Search query string, current admin session
> **Outputs:** Paginated user list (20/page)
> **Constraints:** ≤100ms p95, tenant-scoped, case-insensitive
> **Out of scope:** Fuzzy matching, custom attribute search, audit logging
> **Open questions:** None blocking
>
> 💡 Doctrine candidate: "For any 'add X to dashboard' request, always clarify substring vs fuzzy vs faceted search before planning." Save?

---

*Part of the [Vibe Coding Mastery](https://github.com/AQaddora/vibe-coding-mastery) skill suite. MIT licensed.*

From AQaddora/pure-skill-suite, MIT.

More from AQaddora/pure-skill-suite

Signals

Stars
3
Forks
0
Open issues
0
Last commit (this folder)
May 26, 2026 (5 months ago)
Last release
May 31, 2026 (5 months ago)

From GitHub (fetched Oct 24, 2026).

Details

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

More from AQaddora

Cite this page

understand skill. Agent Mart. https://agentmart.dev/skills/aqaddora-understand (last updated 2026-10-24)