返回全部 Skills

tad-generator

Generate a Technical Architecture Document (TAD) from a PRD. Use when asked to design system architecture or define how a product is built. Updates tad.md and reports GitHub links. Don't use for PRD authoring, sprint tasks, or code implementation.

5

下载量

AI SkillHub 能力展示图

安装方式

命令行安装

在项目根目录执行以下命令,完成 Skill 安装。

npx bzskills add luongnv89/skills --skill tad-generator

skill.md

name: tad-generator
description: Generate a Technical Architecture Document (TAD) from a PRD. Use when asked to design system architecture or define how a product is built. Updates tad.md and reports GitHub links. Don't use for PRD authoring, sprint tasks, or code implementation.
license: MIT
effort: max
metadata:
    version: 1.5.0
    author: "Luong NGUYEN <luongnv89@gmail.com>"

TAD Generator

Generate comprehensive Technical Architecture Documents with modular design for startups.

Subagent Architecture

This skill uses parallel research agents with upfront content extraction. Pattern: D (Research+Synthesis) + E (Staged Pipeline).

Agents

AgentRoleParallelization
prd-reader (agents/prd-reader.md)Read PRD + supporting docs, return structured extractionSequential (only once)
tech-researcher (agents/tech-researcher.md)Handle one research round (spawned 5x in parallel)Parallel (5 instances)
tad-writer (agents/tad-writer.md)Generate complete tad.md from all inputsSequential (after all research)

Research Rounds (5 Parallel)

  • Round 1: Technology Stack validation (React, Node.js, PostgreSQL, Elasticsearch)
  • Round 2: Infrastructure validation (Vercel, AWS, CDN, cost estimation)
  • Round 3: Security review (auth, encryption, compliance, API security)
  • Round 4: Risk assessment (bottlenecks, vendor lock-in, team gaps)
  • Round 5: Holistic review (PRD alignment, team capability, quick wins)

Parallelization Strategy

  • PRD Content: Extracted once by prd-reader, stays out of main context
  • Research Independence: Each round researches conceptually different angle (tech vs. infra vs. security)
  • Reasoning Isolation: Parallel rounds keep each area's reasoning isolated, prevent groupthink
  • Note: Research rounds are conceptual reasoning, not data fetching — parallel rounds won't produce fundamentally different info, but isolation improves quality

Result: All 5 rounds complete concurrently, tad-writer synthesizes outputs into unified TAD.

Environment Check

Before executing:

  1. Verify prd.md exists in project directory
  2. Check for supporting docs (idea.md, validate.md) if available
  3. Confirm WebSearch and WebFetch tools available for research
  4. Verify write permissions to project root for tad.md creation
  5. Ensure git access for final commit

Repo Sync Before Edits (mandatory)

Before creating/updating/deleting files in an existing repository, sync the current branch with remote:

branch="$(git rev-parse --abbrev-ref HEAD)"
git fetch origin
git pull --rebase origin "$branch"

If the working tree is not clean, stash first, sync, then restore:

git stash push -u -m "pre-sync"
branch="$(git rev-parse --abbrev-ref HEAD)"
git fetch origin && git pull --rebase origin "$branch"
git stash pop

If origin is missing, pull is unavailable, or rebase/stash conflicts occur, stop and ask the user before continuing.

Input

Project folder path in $ARGUMENTS containing:

  • prd.md - Product requirements (required)
  • idea.md, validate.md - Additional context (optional)

Workflow

Mode check (do this first): if tad.md already exists in the project folder, this is a Modification run — skip straight to Modification Mode. Otherwise it's a Create run — continue through Phases 1-8 below.

Phase 1: Setup & Validation

  1. Verify prd.md exists
  2. Read supporting docs if present
  3. Read [references/tech-stack.md](references/tech-stack.md) for technology recommendations
  4. Backup existing tad.md if present

Phase 2: Extract Context

From PRD extract:

  • Product name and vision
  • Core features and requirements
  • User flows
  • Non-functional requirements
  • Third-party integrations
  • Analytics requirements

Phase 3: Clarify Architecture

Ask user (if not clear):

DecisionOptions
DeploymentVercel/Netlify (recommended), AWS, GCP, Self-hosted
DatabasePostgreSQL, MongoDB, Supabase/Firebase, Multiple
AuthSocial (OAuth), Email/password, Magic links, Enterprise SSO
BudgetFree tier, <$50/mo, <$200/mo, Flexible

Phase 4: Research & Validation

Spawn one tech-researcher subagent per round, using agents/tech-researcher.md as its prompt, for the 5 research rounds defined under Subagent Architecture above. Run them in parallel; do not reason the rounds inline in main context, which is what the subagent split exists to prevent.

Phase 5: Generate TAD

Create tad.md with sections:

  1. System Overview - Purpose, scope, PRD alignment
  2. Architecture Diagram - Mermaid diagrams for system and flows
  3. Technology Stack - Frontend, backend, database, infrastructure, DevOps
  4. System Components - Modular design with interfaces and dependencies
  5. Data Architecture - Schema, models, flows, storage
  6. Infrastructure - Hosting, environments, scaling, CI/CD, monitoring
  7. Security - Auth, authorization, data protection, API security
  8. Performance - Targets, optimization strategies, caching
  9. Development - Environment setup, project structure, testing, deployment
  10. Risks - Risk matrix with mitigations
  11. Appendix - Research insights, alternatives, costs, glossary

See [references/tad-template.md](references/tad-template.md) for full template structure.

Phase 6: README Maintenance (ideas repo)

After writing tad.md, if the project folder is inside an ideas repo, update the repo README ideas table:

  • Preferred: cd to the ideas repo root and run python3 scripts/update_readme_ideas_index.py if that repo ships it — the script belongs to the user's ideas repo, not to this skill
  • Fallback: update README.md manually (ensure TAD status becomes ✅ for that idea)

Phase 7: Commit and push

  • Commit immediately after updates.
  • Confirm before pushing — this is a visible action:
git push origin <branch>
  • If push is rejected: rebase against the actual upstream tracking branch and retry: branch="$(git rev-parse --abbrev-ref HEAD)"; git fetch origin && git rebase "origin/$branch" && git push.

Phase 8: Output

  1. Confirm tad.md is written (Phase 5) and committed (Phase 7) — do not re-write it here
  2. Summarize architecture decisions
  3. Highlight modular design benefits
  4. List cost estimates by phase
  5. Suggest next steps (setup dev environment, create tasks)

Reporting with GitHub links (mandatory)

When reporting completion, include:

  • GitHub link to tad.md
  • GitHub link to README.md when it was updated
  • Commit hash

Link format (derive <owner>/<repo> from git remote get-url origin):

  • https://github.com/<owner>/<repo>/blob/main/<relative-path>

Step Completion Reports

After completing each major step, output a status report in this format:

◆ [Step Name] ([step N of M] — [context])
··································································
  [Check 1]:          √ pass
  [Check 2]:          √ pass (note if relevant)
  [Check 3]:          × fail — [reason]
  [Check 4]:          √ pass
  [Criteria]:         √ N/M met
  ____________________________
  Result:             PASS | FAIL | PARTIAL

Adapt the check names to match what the step actually validates. Use for pass, × for fail, and to add brief context. The "Criteria" line summarizes how many acceptance criteria were met. The "Result" line gives the overall verdict.

See [references/step-completion-reports.md](references/step-completion-reports.md) for worked examples per phase (Setup, Research, Generation, Output).

Acceptance Criteria

The skill is considered successful when the following are all true. Verify each before reporting completion.

  • [ ] tad.md exists at the project root and contains all 11 required sections (System Overview, Architecture Diagram, Technology Stack, System Components, Data Architecture, Infrastructure, Security, Performance, Development, Risks, Appendix).
  • [ ] Architecture Diagram section contains at least one `mermaid fenced block that parses (no graph typos, balanced braces).
  • [ ] Technology Stack names specific versions or LTS labels for each layer (e.g. Node.js 20 LTS, PostgreSQL 16) — no bare "latest" without a date.
  • [ ] Each item in the Risks section has a paired Mitigation: line (assert one mitigation per risk row).
  • [ ] Infrastructure section lists concrete cost estimates with currency and cadence (e.g. ~$45/mo).
  • [ ] Security section references at least one OWASP control or auth standard (OAuth2, OIDC, JWT, etc.).
  • [ ] Final report includes the GitHub blob URL to tad.md, the commit hash, and (if updated) the README link.
  • [ ] Repo is clean after push: git status returns "nothing to commit, working tree clean".

Expected Output

Example final agent report:

◆ TAD Generation Complete (8 of 8 — delivery)
··································································
  tad.md written:               √ pass (11 sections, 2 mermaid diagrams)
  Versions specified:           √ pass (Node 20 LTS, Postgres 16)
  Risks mitigated:              √ pass (6/6 risks have mitigation)
  Cost estimates:               √ pass (~$45/mo MVP, ~$220/mo scale)
  Committed and pushed:         √ pass (commit a1b2c3d)
  ____________________________
  Result:             PASS

GitHub links:
- tad.md:    https://github.com/acme/my-idea/blob/main/projects/foo/tad.md
- README.md: https://github.com/acme/my-idea/blob/main/README.md
- Commit:    a1b2c3d

Expected tad.md Architecture Diagram excerpt:

## 2. Architecture Diagram

graph TD

U[User] --> CDN[Vercel CDN]

CDN --> APP[Next.js App]

APP --> API[Node API]

API --> DB[(PostgreSQL 16)]

API --> CACHE[(Redis 7)]

Edge Cases

  • Missing prd.md: stop and ask the user to run /prd-generator first; do not invent requirements.
  • PRD too thin (<200 words): warn the user, ask for clarifications on user flows and NFRs before proceeding to Phase 4.
  • Conflicting stack hints in PRD: surface the conflict in Phase 3 clarifying questions; never silently pick one.
  • No git remote origin: skip Phase 7 push, write tad.md locally, and tell the user how to add the remote.
  • Existing tad.md already up to date: enter Modification Mode rather than overwriting; preserve revision history.
  • Non-ideas repo layout: skip Phase 6 README index update; do not create a scripts/update_readme_ideas_index.py if absent.
  • Mermaid render fails locally: validate syntax with mmdc -i tad.md -o /tmp/check.svg (or visual inspection) before commit.

Modification Mode

Triggered by the mode check above when tad.md already exists.

  1. Create a timestamped backup (tad.md.bak.<timestamp>).
  2. Ask which area changed and map the answer to its numbered Phase 5 section: Stack → 3. Technology Stack, Data → 5. Data Architecture, Infrastructure → 6. Infrastructure, Scaling → 6. Infrastructure (scaling is subsection 6.2, not a section of its own), Security → 7. Security.
  3. Apply changes to that section only, preserving the rest of the structure.
  4. Append a revision-history entry (date + summary) at the end of tad.md.

Guidelines

  • Practical: Implementable solutions for startups
  • Cost-conscious: Consider budget implications
  • Modular: Emphasize separation of concerns
  • Specific: Concrete technology choices
  • Visual: Include mermaid diagrams