Skip to content
Back to skills

Sdd Audit

ASecurity

Kiểm định Constitution, Clean Architecture, EARS trace và binding theo Architecture Profile

  • 12 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added September 4, 2026
ai-agentsgosql

Works with

  • cli

Security analysis

A100/100

Scanned September 29, 2026

npx -y skills add khangpqvn/SDDADD --skill sdd-audit --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Sdd Audit?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Sdd Audit
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/khangpqvn-sdd-audit/badge)](https://www.skillsdirectory.com/skills/khangpqvn-sdd-audit)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: sdd-audit
description: Kiểm định Constitution, Clean Architecture, EARS trace và binding theo Architecture Profile
user-invocable: true
---

# SDD Audit (`/sdd-audit`)

**Output language:** All output mirrors the language of the invoking prompt. Vietnamese prompt → Vietnamese output; English prompt → English output. Canonical tokens (`FAIL`, `WARNING`, `PASS`, `CONFIGURATION GAP`, `PENDING HUMAN REVIEW`, `APPROVED`), rule codes (`SEC-01`, `DATA-01`, `ARCH-01`, `ENG-01`), file paths, and CLI commands are language-invariant.

Dùng để kiểm định source và artifact theo `CONSTITUTION.md` trước commit hoặc Pull Request.

## Tham số

- `--feature=<feature-slug>`: Tùy chọn; giới hạn một feature. Không có thì audit toàn repository.

## Architecture Profile

Tuân thủ [Architecture Profile Protocol](../_shared/architecture-profile-protocol.md).

- Audit governance/layer boundary được phép với core-only baseline.
- Route, authentication middleware, ORM/SQL, queue, dependency scan và test command chỉ kiểm tra khi binding tương ứng đã approved/evidenced.
- Binding chưa chọn phải báo `CONFIGURATION GAP`, không kết luận `PASS`/`FAIL` dựa trên framework, SQL dialect, ORM method hay command suy đoán.

## Ba tầng audit

1. **Hard rule:** `SEC-01` secret/PII protection; `SEC-02` identity, authorization ownership và unauthorized behavior cho feature cần access control; `DATA-01` retention, recovery, authorization, audit và deletion policy khi feature đổi core business data.
2. **Architecture:** `ARCH-01` dependency direction và cấm direct DB access từ interface; `ARCH-02` kiểm tra async/reliability behavior theo Spec/profile.
3. **Engineering:** `ENG-01` EARS trace trong `src/usecase/`; `ENG-02` safe error contract theo transport đã approved; `ENG-03` exact approved verification command.

## Kết quả

- `FAIL`: Layer 1 violation hoặc blocker phải sửa.
- `WARNING`: Vấn đề Layer 2/3 cần remediation hoặc rationale approved.
- `PASS`: Check đã chạy và đạt.
- `CONFIGURATION GAP`: Chưa đủ binding/evidence để audit adapter-specific.

## AI Recommendation và Human Final Review

Sau audit, tạo canonical recommendation gồm finding, severity, evidence, remediation option và residual risk. Lưu trong feature artifact hoặc `.sdd/reviews/audit-<slug>.md` với `PENDING HUMAN REVIEW`. Human reviewer có thẩm quyền quyết định disposition; `Tech Lead` có thể review architecture khi repository giao thẩm quyền. Layer 1 failure và blocker còn mở vẫn block. Agent không tự approve audit.

## Hướng dẫn sử dụng
- **Khi dùng:** Read-only audit Constitution, architecture, EARS trace and approved binding evidence before delivery.
- **Không dùng:** Không dùng để self-remediate, edit source/artifact, claim `PASS` for unrun checks, or grant execution.
- **Input:** Optional `--feature`; reject framework-specific conclusions when binding is not approved.
- **Điều kiện trước:** Target repository/feature and profile are readable; adapter checks require matching approved binding and command.
- **Evidence:** Audit recommendation/report with finding severity, path/line evidence, `CONFIGURATION GAP` and residual risk.
- **Dừng khi:** Layer 1 violation, configuration gap, missing exact command, or unresolved finding requiring behavior/governance decision.
- **Human quyết định:** Human disposes findings; code remediation is proposed only and writes occur through approved `/add-execute` task.
- **Lệnh tiếp theo:** `/sdd-review --target=<audit-report> --status=<APPROVED|REVISE|REJECTED> ...`; use `/sdd-update` when a requirement must change.
- **Ví dụ:** `/sdd-audit --feature=feat-orders` produces findings without editing `src/`.

## Completion output

Dùng [Completion output contract](../_shared/ai-review-protocol.md#completion-output-contract). Nêu audit scope, profile/binding state, `PASS`/`WARNING`/`FAIL`/`CONFIGURATION GAP`, report path và residual risk. Kết quả chỉ `PASS` không có warning/finding mới vẫn là `Human decision required` khi recommendation audit còn `PENDING HUMAN REVIEW`; Human reviewer có thẩm quyền ghi decision, reviewer identity và Follow-up không phải placeholder bằng `/sdd-review --target=<feature artifact hoặc .sdd/reviews/audit-<slug>.md> --status=<APPROVED|REVISE|REJECTED> --decision="<human decision>" --reviewer="<authorized human reviewer>" --follow-up="<exact next command or required action>"`. Chỉ follow-up đã persist mới chọn downstream route. Findings thay đổi behavior/governance/profile/contract là `Human decision required` với target review/update phù hợp; Layer 1 `FAIL` hoặc configuration gap là `BLOCKED`. Không claim `PASS` cho check chưa chạy và không tự remediation/bypass gate.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.