Docs Writer

Generate structured documents through guided discovery. 8 document types, each with its own workflow depth.

Workflow

trigger --> detect type --> load reference --> discovery --> drafting

Detect document type from the trigger. If ambiguous, ask the user which type they want.

Document Types

Type	Workflow	Reference	Template
PRD	discovery -> validation -> synthesis -> drafting	prd.md	prd.md
Brief	generated with PRD (no separate trigger)	brief.md	brief.md
Issue	classification -> [clarification] -> drafting	issue.md	issue.md
Task	direct drafting	task.md	task.md
User Story	[clarification] -> drafting	user-story.md	user-story.md
RFC	[clarification] -> analysis -> drafting	rfc.md	rfc.md
ADR	[clarification] -> drafting	adr.md	adr.md
TDD	discovery -> analysis -> drafting	tdd.md	tdd.md

Context Loading Strategy

Load the reference and template matching the detected document type. For types that require discovery, also load discovery.md.

Never simultaneous:

• Multiple document type references
• Templates for different document types

Triggers

Trigger Pattern	Type	Reference
Create PRD, define product, product requirements, write PRD	PRD	prd.md
Create issue, report bug, feature request, create discussion	Issue	issue.md
Create task, new task	Task	task.md
Create user story, write story, new story	User Story	user-story.md
Create RFC, propose change, request for comments	RFC	rfc.md
Create ADR, record decision, architecture decision	ADR	adr.md
Create TDD, technical design, design document	TDD	tdd.md
Create document, write doc	Ask user	--

Shared Discovery Patterns

LOAD: discovery.md before starting any type that requires discovery.

Discovery applies to: PRD, TDD. Clarification (lightweight, only when input is incomplete) applies to: RFC, ADR, Issue (feature/discussion subtypes), User Story. Neither applies to: Task, Issue (bug subtype -- uses structured collection instead), Brief (generated as part of PRD workflow).

Output

All documents save to .artifacts/docs/. Create the directory if it doesn't exist.

Type	Filename
PRD	prd.md
Brief	brief.md
Issue	issue.md
Task	task.md
User Story	story.md
RFC	rfc.md
ADR	adr-{number}-{name}.md
TDD	tdd.md

For ADR, use kebab-case for {name} and auto-detect the next sequential number from existing files in .artifacts/docs/.

Quality Standards

Requirements must be concrete and measurable across all document types.

Bad	Good
"Search should be fast"	"Search returns results within 200ms"
"Easy to use"	"New users complete onboarding in under 2 minutes"
"Intuitive interface"	"Task completion rate above 90% without help text"

Cross-References

PRD -------------> TDD            (PRD feeds requirements, journeys, rules into TDD)
PRD -------------> design-builder (PRD + Brief inform copy and design extraction)
PRD -------------> spec-driven    (PRD milestones feed spec initialization)
RFC -------------> ADR            (accepted RFC generates ADR)
TDD -------------> ADR            (references relevant decisions)

Notes:

• design-builder: PRD sections 1, 3-4 (problem, personas, scope) and Brief (value prop, market) inform copy extraction and design extraction
• spec-driven: PRD milestones feed feature initialization -- each milestone can generate a spec with its own tasks

Guidelines

DO:

• Always complete discovery before drafting (for types that require it)
• Present draft for user feedback before saving
• Mark unknowns as TBD rather than inventing constraints
• Use concrete, measurable requirements
• Use fixed filenames per type (ADR keeps {name} suffix in kebab-case)

DON'T:

• Skip discovery for types that require it
• Assume document type -- detect from trigger or ask
• Include visual/design direction (that belongs in design-builder)
• Use vague adjectives as requirements ("fast", "easy", "intuitive")
• Mix document types in a single file

Error Handling

• No .artifacts/docs/: Create the directory
• Ambiguous trigger: Ask user which document type
• Missing context for discovery: Ask questions, never assume
• ADR numbering conflict: Scan existing files and use next available number