About tech-review-doc
A skill for turning code-heavy design documents into technical reviews for product, QA, and readers who are not close to the code. It clarifies goals first, then creates needed diagrams and sections, with explicit user approval required before proceeding past §1.
Key features
- 01Generate reviews from a template
- 02Clarify and confirm business goals first
- 03Create necessary Mermaid diagrams
- 04Compare options and summarize core changes as needed
- 05Enforce an approval gate and output rules
Use cases
- 01Turn implementation designs into reviews for product and QA
- 02Confirm goals and acceptance focus before writing a review
- 03Extract module changes, data flows, and risks from complex designs
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ fudesign2008/ open-skills --skill tech-review-docAfter installing, type/tech-review-docin Claude Code to confirm it loaded.
Bundled filestemplate.md
Details
From the project READMETechnical Design → Technical Review Document
Convert a code-heavy design document into a technical review document for product, QA, and engineers who are not deep in the code.
Core principles
| Principle | Requirement |
|---|---|
| Audience | Non-code-deep readers can understand §§1–4 on their own |
| Language | Concise and plain; §§1–3 zero code; from §4 onward keep only necessary interface/field/table purpose |
| Diagrams | Mermaid; nodes use business/module names, not class or file names |
| Source of truth | The design doc is factual; user-confirmed content overrides vague design wording |
| Brevity | For each section ask: “Does anyone in the review meeting need this?” — if not, omit it |
| Template | Use sibling template.md |
Inputs
| Input | Notes |
|---|---|
| Design document path (required) | User @ or path |
| Review type (optional) | Scheme / implementation / acceptance review; default scheme review |
Read the design document fully before starting; scan related PRD or OpenSpec change when useful.
If the design exceeds ~1500 lines, skim structure and headings first, then deep-read sections as needed.
Output
Write the file next to the input document, named tech-review-{topic}-{YYYY-MM-DD}.md.
Derive {topic} from the design title keywords; use the current date for YYYY-MM-DD.
Flow (five steps, hard gate)
Step 0 Startup check
↓
Step 1 Background & goals (clarify; no file write until user approves §1)
↓ user explicitly approves §1
Step 2 Generate §2 necessary diagrams (ask if unsure)
↓
Step 3 Multi-option comparison → §3 (skip if none)
↓
Step 4 Core content summary → §4
↓
Step 5 Complete §§5–7 and write the file
**Until Step 1 is done and the user explicitly approves §1, do not create or write the review file, and do not run Steps 2–5.**
Prerequisite skill check
On load, verify clarifying-question-discipline is available. If missing, abort and print:
npx skills add FuDesign2008/open-skills -g --skill clarifying-question-discipline --yes
Do not silently fall back.
Step 0: Startup check
- Confirm the design document path; ask if missing or invalid
- Read sibling
template.md - Extract from the design: title, surfaces/modules, whether multi-option comparison exists, existing Mermaid
- Restate understanding in 3–5 sentences and state that Step 1 clarification starts next
Step 1: Background & goals (mandatory)
Design backgrounds are usually written for engineers. Review §1 must use business language for product/QA.
Follow clarifying-question-discipline: one question per turn, multi-round until clear; prefer structured single-choice when the agent supports it; include a recommended answer when a default is knowable. Describe intent only — do not hard-code a platform-specific questioning tool.
1.1 Goal
Jointly confirm and refine §1 Background & goals. Treat design-doc background as a draft only — never paste it verbatim.
1.2 Coverage (ask across rounds as needed)
Business motivation, user scenarios, success criteria, Non-Goals, glossary terms that product/QA need.
Translate technical motives into business language before asking for confirmation.
1.3 Suggested questions (pick one at a time)
- Whose problem, and what problem, does this mainly solve?
- What is the business cost of doing nothing?
- What change can users/ops directly perceive?
- What is the single most important QA acceptance check?
- What is explicitly out of scope?
- Do any terms need more business-friendly names for product/QA?
1.4 Deliverable and gate
- Present a §1 draft (business language only, no code)
- Ask: “Is this background and goal accurate? After you confirm, I will generate diagrams and the full document.”
- Enter Step 2 only after explicit user confirmation
Step 2: §2 Solution overview (diagrams)
2.1 Which diagrams (as needed)
| Diagram | When needed |
|---|---|
| System topology | ≥2 clients/deploy boundaries |
| Architecture | Module responsibilities need a separate view |
| Data flow | Config/state crosses modules, or migration |
| Logic flow | Main path has branches, validation, or failure paths |
Ask the user when unsure. Only draw diagrams that are needed; omit unused diagram types from the document.
2.2 Drawing rules
- Rewrite Mermaid from the design; do not copy diagrams that embed code symbols
- Each diagram: title + 1–2 sentence caption tying back to §1 goals
- §2.1 solution summary: 2–3 sentences
Step 3: §3 Options and selection (conditional)
| Case | Action |
|---|---|
| Substantive multi-option comparison exists | Generate §3 (decision context + comparison table + selection conclusion only) |
| Single chosen approach, no comparison | Skip §3 |
Comparison dimensions: goal fit, user experience, delivery cost, risk/rollback, long-term maintenance.
Step 4: §4 Core content summary
Keep
- Change scope (product module names)
- Main flows; meaning of critical data/config; storage and migration
- Business meaning of interface changes (fold into flow narrative; no standalone API tables)
Drop or compress
- Function signatures, class names, large code/JSON dumps
- Unit-test filenames, commit-level churn
- Standalone interface/contract tables
Self-check: a developer unfamiliar with the business can answer “which areas change, how data moves, what happens to old data” after §4.
Step 5: §§5–7 and write file
| Section | Points |
|---|---|
| §5 Risks & boundaries | Risk list (required) + known limits (required); security/compliance only when relevant |
| §6 Release & ops | Conditional: only when canary/rollout or server-side data backfill applies; otherwise omit |
| §7 Conclusion | Draft verdict + Action Items |
After writing, remove the template’s trailing “模板使用说明” section. In chat, report: output path, whether §1 was confirmed, which diagrams were generated, whether §3/§6 were skipped, and any placeholders left for the user.
Language conversion reference
| Technical phrasing | Review phrasing |
|---|---|
| Call API / RPC | A requests data from B |
| Write Redis / cache | Temporarily remember this state for next time |
| DB migration | Reorganize old data under the new rules |
| Config push | Admin rules sync to clients |
| Degrade / fallback | On failure, automatically fall back to the old path |
| Idempotent | Repeating the action has no extra side effects |
| Transaction | All succeed together, or nothing changes |
Prefer “who does what to whom, and what the user sees” over implementation detail.
Host auto-mode contract
When invoked under write-workflow auto mode:
- The §1 HARD-GATE still applies — no file write and no Steps 2–5 until the user explicitly approves §1.
- After §1 approval, Steps 2–5 MAY run continuously (diagrams → remaining sections → write file) without per-step confirmation.
- Standalone or manual-host invocation keeps the interactive §1 gate as in Step 1.4.
Relationship to other skills
| Skill | Relationship |
|---|---|
clarifying-question-discipline | Strong dependency for Step 1 |
write-workflow | Optional host (eight-stage) that routes here by default for tech-review writing |
Success criteria
- Step 1 explicitly confirmed by the user
- §§1–3 have no code blocks except Mermaid
- §2 only necessary diagrams
- §3 generated or skipped per rules
- §6 generated or skipped per rules
- §4 is not a code dump
- File written with traceable link to the design doc
- Path and naming match the Output rules