Technical Documentation Skill

Create professional technical documentation using structured templates with interactive guidance. This skill helps users articulate their ideas clearly, producing well-organized documents.

Key principle: All information comes from the user through interactive prompts. Do NOT read the user's codebase to discover or understand the problem. The Read tool is ONLY for reading user-provided context files (e.g., existing ADRs, RFCs, design documents) that the user explicitly references for additional context.

Skill Philosophy

This skill follows a guided discovery approach rather than automatic generation:

• Information comes from the user, not the codebase
• Progressive questioning reveals user's thinking
• Templates structure the thought process
• Research assistance when needed, not automatic
• Quality validation ensures completeness

This approach produces better documentation because it forces users to think through their proposals systematically.

Supported Document Types

Type	Template	Status	Use Case
One-Pager	assets/templates/one-pager/	✅ Active	Proposals, feature briefs, quick decisions
RFC	assets/templates/rfc/	✅ Active	Request for Comments, cross-team design proposals
TSD	assets/templates/tsd/	✅ Active	Technical Specification Documents (APIs, data models)
ADR	assets/templates/adr/	✅ Active	Architecture Decision Records (technology choices)
POC/Experiment	assets/templates/poc-experiment/	✅ Active	Proof of concepts, technical experiments, technology validation

Three-Phase Architecture

IMPORTANT: This skill now follows a unified three-phase architecture that applies to ALL document types:

1. Phase 1: Universal Discovery (Problem Validation + Context Input) - SAME FOR ALL DOCUMENT TYPES
2. Phase 2: Document Type Selection (AI recommends, user decides)
3. Phase 3: Template-Specific Discovery (Questions unique to chosen document type)

Key Benefits:

• Problem clarity BEFORE committing to document format
• AI can recommend the most appropriate format based on the situation
• Single unified requirements gathering works for all document types
• Template-specific questions are more focused (problem already understood)

Workflow Overview

IMPORTANT: You MUST follow the workflow gates below. Do NOT skip to content generation without completing mandatory steps.

Workflow Gates (Mandatory Checkpoints)

Gate	Checkpoint	Required?	Blocked Until
1	Problem Validation Gate complete	✅ Yes	Cannot proceed without validated problem
2	Document type selected	✅ Yes	Cannot proceed without type selection
3	Research offered (if Exploratory/Research-first)	Conditional	Must offer before content if user is uncertain
4	Quality validation	✅ Yes	Cannot present final output without validation

Phase 1: Universal Discovery (ALL Document Types)

CRITICAL: Phase 1 is MANDATORY for ALL document types. The Problem Validation Gate must be completed BEFORE selecting a document type.

Goal: Validate the problem and gather rich context BEFORE deciding on document format.

Step 1: Readiness Check

Ask user to choose their readiness level. See references/question-bank.md for the AskUserQuestion pattern.

Readiness	Description	Workflow
Quick Start	User has all details ready	Skip to Phase 2 (Document Type Selection)
Guided	User has basics, needs help	Complete Problem Validation Gate → Phase 2
Exploratory	Just an idea	Problem Validation Gate → Research (to fill gaps) → Phase 2
Research-first	Needs to gather information	Problem Validation Gate → Research (to gather context) → Phase 2

Step 2: Problem Validation Gate (MANDATORY)

GATE 1: You MUST complete the Problem Validation Gate BEFORE selecting a document type or discussing solutions.

For Guided, Exploratory, and Research-first modes, complete the Universal Discovery questions:

Mandatory Problem Validation (4 questions):

1. Current State Deep Dive: "Walk me through what happens now step by step."
2. Evidence of Problem: "How do we know this is a problem? What evidence exists?"
3. Who Has This Problem (conditional): If business-facing → "Who specifically experiences this?"; If technical-only → "Which teams/services are affected?"
4. Current vs Desired State: "Complete these sentences: 'Today, ________ happens.' and 'After we fix this, ________ should happen.'"

Enhanced Problem Questions (optional): 5. Pain Point Specificity: "What's the SINGLE WORST thing about the current situation?" 6. Impact Urgency: "What's the cost of NOT fixing this?"

Solution Challenge Flow: If user jumps to solution before completing Problem Validation Gate:

• Acknowledge the solution idea
• Ask: "What PROBLEM does this solve?"
• Complete Problem Validation Gate
• Then proceed to Phase 2

Step 3: Rich Context Input (Optional)

These can be asked at ANY point to enrich the proposal:

• Related Documents: "Are there related documents we should reference? (ADRs, RFCs, TSDs, tickets)"
• Document Attachment Flow: "You can provide context in three ways: File paths (I'll read them), Paste content, Verbal description. Which do you prefer?"
• Historical Context: "Has this been proposed or attempted before?"
• Technical Context: "What technologies/systems are involved? Any constraints?"
• Team Context: "Who will implement this? Any organizational constraints?"

See references/question-bank.md for Universal Discovery question patterns (see "Universal Discovery Questions" section).

Phase 2: Document Type Selection (NEW)

GATE 2: You MUST select a document type AFTER completing Universal Discovery (or after Quick Start users have all information).

Goal: Recommend and select the most appropriate document type based on the validated problem.

Step 1: AI Analysis

Analyze the problem to recommend a document type:

Decision Logic:

Indicators	Recommended Type	Rationale
Simple problem, single solution, quick decision	One-Pager	Concise proposal for fast approval
Complex architecture, multiple services, implementation phases	RFC	Detailed design requires thorough analysis
API specification, data model, interface definition	TSD	Technical implementation details
Technology choice, framework decision, architectural pivot	ADR	Decision-focused with alternatives analysis
Hypothesis validation, technology evaluation, uncertain feasibility	POC/Experiment	Learning-focused with Go/No-Go decision
User unsure, unclear scope	Ask User	Present options with descriptions

Step 2: Present Recommendation

Example:

Based on what you've described, I recommend an **RFC** because:
- This involves multiple services and implementation phases
- There are significant migration and rollback considerations
- The architectural approach needs thorough review

Does that sound right?

Step 3: User Decision

Options:

• Accept AI recommendation
• Choose different type (with explanation)
• See all options with descriptions
• Ask for help deciding

Phase 3: Template-Specific Discovery

Goal: Ask questions unique to the chosen document type AFTER the problem is understood and type is selected.

Process:

1. Load the appropriate template guidance: assets/templates/{type}/guidance.md
2. For each section, ask template-specific questions from guidance.md
3. Use Universal Discovery output in applicable sections (see guidance.md for mapping)
4. Generate content using template structure

Template-Specific Questions Examples:

• One-Pager: Sign-offs required, timeline urgency, high-level risks
• RFC: Architecture overview, implementation phases, migration strategy, rollback plan
• TSD: API endpoints, data models, interface contracts, error handling
• ADR: Alternatives considered (mandatory 2+), decision criteria, trade-offs analysis
• POC/Experiment: Hypothesis statement, success criteria, measurement approach, Go/No-Go criteria

Research Integration (within Phases)

When to offer research:

• During Phase 1 (Universal Discovery): AFTER Problem Validation Gate, if user selects "Exploratory" or "Research-first" mode
  • Research fills gaps identified during Problem Validation (e.g., "we don't have metrics on X", "we need to know what competitors do")
• During Phase 3 (Template-Specific): If template-specific questions reveal information gaps

How to invoke research:

• Invoke agent:web-research-specialist for industry best practices, technical context, competitor analysis
• Use research findings to enrich Universal Discovery output or answer Template-Specific questions

Key Principle: Problem Validation Gate ALWAYS comes first. Research is targeted based on what gaps are identified during validation.

Phase 4: Content Generation

CRITICAL: Output Rules

⚠️ What to INCLUDE in user-facing documents:

• ✅ Section headings from template.md (e.g., ## Problem, ## High level goal)
• ✅ User-provided content and information
• ✅ Diagrams (if requested and generated via mermaid skill)
• ✅ Placeholder text in italics (e.g., Describe the problem)
• ✅ Tables, lists, and formatting for user content

⚠️ What to EXCLUDE from user-facing documents:

• ❌ Checkboxes (✅ or - [ ]) of any kind
• ❌ HTML comments (<!--- ... --->)
• ❌ Quality criteria statements
• ❌ "GUIDANCE QUESTIONS" sections
• ❌ References to quality-checklist.md
• ❌ Any validation or assessment language

Template Files Purpose:

• template.md - Clean structure with section headings (use for document structure)
• guidance.md - Section-specific guidance with questions and quality criteria (read for context, never copy to output)
• quality-checklist.md - Validation criteria (use in Phase 5 only, never show to user)

Content Generation Process:

1. Load template guidance: Read the appropriate assets/templates/{type}/guidance.md
2. Work section-by-section: For each template section:
   • Explain what the section needs
   • Ask targeted questions from guidance.md (use choices where possible)
   • Generate draft content WITHOUT any checkboxes or quality criteria
   • Get feedback before moving on
3. Reference examples: Use assets/templates/{type}/examples.md for quality benchmarks

Visual Aids (Optional)

Only if user requested diagrams:

1. Ask what the diagram should show
2. Use common-engineering:mermaid skill to generate validated diagrams
3. Integrate into document

Do NOT proactively suggest diagrams unless user requested them.

Phase 5: Review & Output

MANDATORY: Pre-Output Self-Check

Before presenting ANY draft to the user, you MUST verify:

• ❌ NO checkboxes (✅ or - [ ]) anywhere in the document
• ❌ NO HTML comments in the document
• ❌ NO quality criteria text in the document
• ❌ NO "GUIDANCE QUESTIONS" sections in the document
• ❌ NO references to quality-checklist.md in the document
• ✅ ONLY section headings and user-provided content included

If ANY prohibited content found: Remove it immediately before presenting to user.

Phase 5 Workflow:

1. Present Draft (WITHOUT validation criteria):

   • Show the complete document to the user
   • Do NOT include any checkboxes, quality criteria, or validation language
   • The draft should contain ONLY section headings and user content

2. Internal Validation (Using quality-checklist.md):

   • Read assets/templates/{type}/quality-checklist.md
   • INTERNALLY validate the document against quality criteria
   • Do NOT show the checklist to the user
   • Do NOT include checkboxes in any response

3. Offer Improvements (Without showing checklist):

   • If gaps identified during validation, ask targeted questions
   • Suggest enhancements based on quality criteria (but don't show the criteria)
   • Example: "Would you like to add more specific metrics to the goal section?" (NOT "The checklist says...")

4. User Refinements:

   • Let user request changes
   • Make updates as requested
   • Continue to ensure no prohibited content appears

5. Final Output:

   • Run Pre-Output Self-Check one more time
   • Save using document-skills:docx or document-skills:pdf
   • Confirm no prohibited content in final document

How to Invoke This Skill

This skill activates automatically when users request technical documentation. It works through interactive prompts:

1. User trigger: User says "write a one-pager" or similar phrase
2. Skill activation: techdocs skill loads and begins workflow
3. Interactive prompting: Skill asks questions using AskUserQuestion
4. Research integration: If needed, invoke agent:web-research-specialist using Task tool
5. Diagram generation: If needed, invoke common-engineering:mermaid skill

Integration Examples

Research assistance:

When user selects "Research-first" or input is vague:
1. Use Task tool with subagent_type="common-engineering:web-research-specialist"
2. Pass research query (e.g., "industry best practices for OAuth implementation")
3. Incorporate findings into document sections

Diagram creation:

When user requests diagrams:
1. Invoke common-engineering:mermaid skill
2. Provide diagram type and content requirements
3. Mermaid skill validates and generates PNG
4. Include diagram in document output

Quality Validation

⚠️ CRITICAL: Quality validation is for INTERNAL use only. Never show checkboxes or validation criteria to users.

How to validate:

1. Read the validation file: assets/templates/{type}/quality-checklist.md
2. Validate internally: Check the document against criteria WITHOUT showing the checklist
3. Address gaps: If issues found, ask targeted questions to improve content
4. Never show criteria: Offer improvements without mentioning the validation checklist

Example - CORRECT approach:

• ✅ "Would you like to add specific metrics to measure success?"
• ✅ "I notice the risks section could use mitigation strategies. Shall I help with that?"

Example - WRONG approach:

• ❌ "The quality checklist says we need metrics"
• ❌ "Let me check if this passes validation: - [ ] Metrics included"

Universal Validation Criteria (Internal Use Only)

These criteria are checked internally during Phase 5:

• All required sections complete
• Metadata filled (authors, date, status)
• Problem clearly stated with impact
• Clear recommendation made
• Risks identified with mitigations

Document-Specific Validation

Each document type has detailed validation criteria in assets/templates/{type}/quality-checklist.md. Read this file during Phase 5 but NEVER show it to users.

Reference Files

File	Purpose	Use In
writing-guidelines.md	Technical writing best practices	All phases
question-bank.md	Reusable AskUserQuestion patterns	Phase 1, 3
assets/templates/{type}/template.md	Clean document structure (headings only)	Phase 3 (structure)
assets/templates/{type}/guidance.md	Section-specific guidance with questions and quality criteria (DO NOT copy to output)	Phase 3 (read for context)
assets/templates/{type}/quality-checklist.md	Validation criteria (DO NOT show to user)	Phase 5 (internal validation)
assets/templates/{type}/examples.md	Completed example documents	Phase 3 (quality reference)

File Usage Rules:

• ✅ USE template.md for document structure
• ✅ READ guidance.md for understanding
• ✅ READ quality-checklist.md for internal validation
• ❌ DO NOT copy content from guidance files to user output
• ❌ DO NOT show checkboxes or quality criteria to users

Integration Points

Required Skills/Agents

• agent:web-research-specialist (common-engineering plugin)

  • Purpose: Research industry best practices, technical context, competitors
  • When to use: User selects "Research-first" or "Exploratory" mode
  • Invocation: Task(subagent_type="common-engineering:web-research-specialist", ...)

• common-engineering:mermaid (common-engineering plugin)

  • Purpose: Generate validated Mermaid diagrams with automatic syntax checking
  • When to use: User requests diagrams or you recommend them for complex flows
  • Invocation: Invoke skill via "create a [diagram-type] diagram showing [content]"

Optional Skills (Output)

• document-skills:docx (document-skills plugin)

  • Purpose: Export document to Microsoft Word format
  • Invocation: Use when user selects "Word" output format

• document-skills:pdf (document-skills plugin)

  • Purpose: Export document to PDF format
  • Invocation: Use when user selects "PDF" output format

Dependencies

⚠️ Required plugins:

• common-engineering (for web-research-specialist agent and mermaid skill)
• document-skills (for docx/pdf export - optional, can output markdown without this)

Adding New Document Types

To add a new document type (e.g., POC/Experiment):

1. Create assets/templates/poc-experiment/ directory
2. Add template.md - Document structure with guidance comments
3. Add guidance.md - Section-by-section guidance for the agent
4. Add quality-checklist.md - Validation criteria (internal use only)
5. Add examples.md - 2-3 completed examples
6. Update the document types table above (line 29-37)
7. Update techdocs-writer.md agent Decision Logic table
8. Update techdocs-writer.md agent Template-Specific Questions Examples
9. Add any type-specific questions to references/question-bank.md

The core workflow in this skill handles all document types generically. Document-specific logic lives in the template files.

Adaptive Modes

The three-phase architecture supports different user readiness levels:

Quick Start Mode

For users with "I have everything ready":

1. Skip to Phase 2 (Document Type Selection)
2. User provides problem context directly
3. AI recommends document type based on user input
4. Complete Phase 3 (Template-Specific Discovery)
5. Generate content using template
6. Review and output

Guided Mode

For users with "I have the basics":

1. Phase 1: Complete Problem Validation Gate
2. Phase 2: AI recommends document type
3. Phase 3: Template-specific questions for each section
4. Generate content section-by-section with feedback
5. Review complete document
6. Output

Exploratory Mode

For users with "Just an idea":

1. Phase 1: Problem Validation Gate → Research (to fill gaps)
2. Phase 2: AI recommends document type based on validated problem + research findings
3. Phase 3: Template-specific discovery
4. Content generation with research findings integrated
5. Review and output

Research-First Mode

For users who need to gather information:

1. Phase 1: Problem Validation Gate → Research (to gather context) → Context Input
2. Phase 2: AI recommends document type
3. Phase 3: Template-specific discovery
4. Content generation with research findings
5. Review and output

Best Practices

1. Always assess readiness first: Before generating ANY content, ask how much information the user has ready
2. Offer research proactively: If user input is vague or they seem uncertain, offer research assistance
3. Ask before assuming: Confirm understanding before generating content
4. Use choices: Prefer multiple-choice questions over open-ended when possible
5. Skip what's known: Don't re-ask information user already provided
6. Show progress: Let user know which section you're working on
7. Validate quality: Use the checklist before presenting final output
8. Reference examples: Point users to examples.md when they're unsure of format