decebal Ralph TUI PRD Generator (chronis)
Create detailed Product Requirements Documents optimized for AI agent execution via ralph-tui with the chronis (cn) CLI.
The Job
- Receive a feature description from the user
- Ask 3-5 essential clarifying questions (with lettered options) - one set at a time
- Always ask about quality gates — classify into epic-level vs. story-level
- After each answer, ask follow-up questions if needed (adaptive exploration)
- Generate a structured PRD when you have enough context
- Output the PRD wrapped in
[PRD]...[/PRD]markers for TUI parsing
Important: Do NOT start implementing. Just create the PRD.
Step 1: Clarifying Questions (Iterative)
Ask questions one set at a time. Each answer should inform your next questions. Focus on:
- Problem/Goal: What problem does this solve?
- Core Functionality: What are the key actions?
- Scope/Boundaries: What should it NOT do?
- Success Criteria: How do we know it's done?
- Integration: How does it fit with existing features?
- Quality Gates: What commands validate success, and which tier? (REQUIRED)
- Testing Trophy: What test layers are needed? (unit, contract, integration, frontend, E2E) (REQUIRED)
Format Questions Like This:
1. What is the primary goal of this feature?
A. Improve user onboarding experience
B. Increase user retention
C. Reduce support burden
D. Other: [please specify]
2. Who is the target user?
A. New users only
B. Existing users only
C. All users
D. Admin users only
This lets users respond with "1A, 2C" for quick iteration.
Quality Gates Question (REQUIRED)
Always ask about quality gates with two-tier classification:
What commands validate the codebase? These run ONCE when all stories are done (epic-level):
A. task ci (go-task runner)
B. make ci
C. bun typecheck && bun lint
D. cargo test
E. Other: [specify your commands]
For individual stories, what story-specific checks apply?
A. UI stories: verify in browser using dev-browser skill
B. UI stories: Playwright e2e test for that story
C. Backend stories: curl/test the specific endpoint
D. Other: [specify]
Why two tiers? Running bun typecheck after every single story wastes agent context and time. Intermediate stories may have temporary type errors that the next story fixes. General checks run once at the end.
Adaptive Questioning
After each response, decide whether to:
- Ask follow-up questions (if answers reveal complexity)
- Ask about a new aspect (if current area is clear)
- Generate the PRD (if you have enough context)
Typically 2-4 rounds of questions are needed.
Testing Trophy Question (REQUIRED)
Always ask about test coverage expectations:
What test layers does this epic need?
A. Unit only (pure logic, projections, state machines)
B. Unit + Contract (data shapes across IPC boundary)
C. Unit + Contract + Integration (cross-service flows)
D. Full trophy: Unit + Contract + Integration + Frontend + E2E
E. Other: [specify]
Which behaviors are "block merge" critical?
A. Auto-completion / background logic (bugs are invisible to users)
B. Data filtering / query correctness (wrong data shown)
C. Orchestration (wrong things triggered)
D. All of the above
E. Other: [specify]
Use the answers to populate the Testing Trophy section in the PRD.
Step 2: PRD Structure
Generate the PRD with these sections:
1. Introduction/Overview
Brief description of the feature and the problem it solves.
2. Goals
Specific, measurable objectives (bullet list).
3. Quality Gates (Two Tiers)
CRITICAL: Split gates into two tiers. The conversion tool (ralph-tui-cn-beads) reads this section directly.
## Quality Gates
### Epic-Level (run once on epic completion)
General codebase checks. These validate the whole project and run ONCE when all stories are done:
- `task ci` — full CI pipeline (typecheck + lint + test)
- `bun typecheck` — type checking
- `bun lint` — linting
### Story-Level (checked per story)
Criteria checked on specific stories where relevant:
- **UI stories:** Verify in browser using dev-browser skill
- **Backend stories:** Verify endpoint returns expected response
- **Migration stories:** Verify column/table exists after migration
4. User Stories
Each story needs:
- Title: Short descriptive name
- Type tag:
[Schema],[Backend],[UI],[Integration]— tells the conversion tool which story-level gates apply - Description: "As a [user], I want [feature] so that [benefit]"
- Acceptance Criteria: Verifiable
- [ ]checkboxes the agent marks as done - Mark-each-item instruction: Reminder to verify and check off each item
Each story should be small enough to implement in one focused AI agent session.
Format:
### US-001: [Title] [Schema]
**Description:** As a [user], I want [feature] so that [benefit].
**Acceptance Criteria:**
- [ ] Specific verifiable criterion
- [ ] Another criterion
- [ ] (story-level gate if applicable)
Mark each item [x] as you complete it. Only close when all are checked.
Important — writing verifiable criteria:
Every criterion must be something the agent can concretely verify:
Good (agent can check these):
- [ ] File apps/core/src/new_module.rs exists and compiles- [ ] GET /api/v1/events returns 200 with JSON body- [ ] Column investorType added with default 'cold' (check migration file)- [ ] Component renders without errors (verify in browser)
Bad (agent cannot verify these):
- [ ] Works correctly- [ ] Good UX- [ ] Handles edge cases- [ ] Performant
Note: Do NOT include epic-level gates (typecheck, lint, ci) in individual story criteria — they are defined once in the Quality Gates section and run on epic completion only. Story-level gates (browser verification, endpoint testing) DO belong in the relevant stories.
5. Testing Trophy
REQUIRED for every epic. Define the expected test coverage across all layers BEFORE implementation begins. This prevents shipping features with test debt.
The testing trophy prioritizes integration and contract tests over unit tests and E2E:
╭───────╮
│ E2E │ Few, slow, high confidence
╭──┴───────┴──╮
│ Integration │ Cross-boundary flows
╭──┴─────────────┴──╮
│ Contract/Handler │ Data shapes, API boundaries
╭──┴───────────────────┴──╮
│ Unit (logic only) │ Pure functions, state machines
╰──────────────────────────╯
For each layer, list specific tests the epic requires:
## Testing Trophy
### Unit Tests
- [ ] [Projection] TaskRunProjection handles Started/Completed events (N tests)
- [ ] [Store] get_outputs filters deliverable steps only
- [ ] [Logic] auto-completion derives correct status from sibling runs
### Contract Tests
- [ ] [Handler] list_task_runs returns sorted TaskRunSummary array
- [ ] [Handler] run_job returns task_run_id, creates workflows
- [ ] [Handler] get_task_run_outputs filters is_deliverable == true
- [ ] [Frontend API] listTaskRuns/runJob/getTaskRunOutputs match handler shapes
### Integration Tests
- [ ] [Service] run_job creates TaskRun + triggers linked workflows
- [ ] [Flow] workflow completion → auto-completes parent TaskRun
### Frontend Tests
- [ ] [Component] Empty state shows Run Job CTA
- [ ] [Component] Accordion expand loads outputs
- [ ] [Component] Approve/Reject actions invoke correct commands
TDD-First Rules:
Every story follows test-driven development. Tests are written BEFORE or ALONGSIDE implementation, never after.
- Every user story must list its specific tests in the acceptance criteria:
- [ ] Test: [description] (file: [path]) - Backend stories follow red-green-refactor:
- First criterion: write failing test(s) for the expected behavior
- Middle criteria: implement until tests pass
- Last criterion: verify all tests green
- UI stories include contract tests for data shapes:
- [ ] Contract test: [invoke command] returns expected shape - Each story carries its own tests — no "write all tests at the end" stories
- Epic-level quality gates include:
Testing trophy complete — all layers covered per the Testing Trophy section
Test Placement in Stories (examples):
### US-003: Create TaskRun entity [Backend]
**Acceptance Criteria:**
- [ ] Test: TaskRunProjection handles Started event and queries by ID (red first)
- [ ] Test: TaskRunProjection derives status from child workflow runs
- [ ] TaskRun struct with fields: task_run_id, job_id, status, ...
- [ ] AllSource projection registered
- [ ] All tests green: cargo test task_run
### US-007: Expandable accordion [UI]
**Acceptance Criteria:**
- [ ] Contract test: getTaskRunOutputs returns WorkflowRunOutputs[] with nested steps
- [ ] Contract test: approve_step_run accepts stepRunId + disposition
- [ ] Accordion expands on click, loads outputs via getTaskRunOutputs
- [ ] All tests green: bun test job-detail-runs
Classify tests by deployment gate:
- Block merge: Implicit/background logic (auto-completion, data filtering, orchestration) — bugs here are invisible to users
- Block release: Handler contracts, frontend component behavior, review action correctness
- Follow-up OK: E2E, visual regression, performance edge cases
Anti-patterns to reject:
- ❌ Shipping all stories without tests, then adding a "write tests" story at the end
- ❌ Only testing projections (base of trophy) while skipping contract and integration layers
- ❌ "It's just UI" — interactive logic (accordion, review, state machines) needs contract tests
- ❌ Tests that verify implementation details instead of behavior ("function was called" vs "output is correct")
- ❌ Acceptance criteria that say "tests added" without naming the specific tests
6. Functional Requirements
Numbered list of specific functionalities:
- "FR-1: The system must allow users to..."
- "FR-2: When a user clicks X, the system must..."
Be explicit and unambiguous.
7. Non-Goals (Out of Scope)
What this feature will NOT include. Critical for managing scope.
8. Technical Considerations (Optional)
- Known constraints or dependencies
- Integration points with existing systems
- Performance requirements
9. Success Metrics
How will success be measured?
10. Open Questions
Remaining questions or areas needing clarification.
Writing for AI Agents
The PRD will be executed by AI coding agents via ralph-tui. Therefore:
- Be explicit and unambiguous
- User stories should be small (completable in one agent context window)
- Every acceptance criterion must be a
- [ ]checkbox the agent can verify and mark- [x] - Include specific file paths if you know them
- Reference existing code patterns in the project
- Tag stories with type (
[UI],[Schema],[Backend]) so story-level gates are unambiguous - Include "Mark each item [x]..." instruction in every story
Output Format
CRITICAL: Wrap the final PRD in markers for TUI parsing:
[PRD]
# PRD: Feature Name
## Overview
...
## Quality Gates
### Epic-Level (run once on epic completion)
...
### Story-Level (checked per story)
...
## User Stories
...
[/PRD]
File naming: The TUI will save to ./docs/proposals/prd-[feature-name].md
Example Conversation Flow
User: Create a PRD for adding dark mode to the TUI
Assistant: I'll help you create a PRD for dark mode. Let me ask a few questions:
What is the primary goal? A. Reduce eye strain for long sessions B. Match system theme preference C. Provide visual customization D. Other
Should the setting persist? A. Yes, save to config file B. No, session only C. Follow system preference
User: 1A, 2A
Assistant: Great! A couple more questions:
Which components need dark mode support? A. All TUI components B. Only the main execution view C. Main view + settings D. Other
What commands validate the whole codebase? (epic-level, run once at the end) A. task ci B. bun typecheck && bun lint C. make ci D. Other: [specify]
For UI stories, what story-specific checks? A. Verify in browser using dev-browser skill B. Playwright e2e test per story C. Both A and B D. No story-level checks needed
User: 3A, 4B, 5A
Assistant: Perfect, I have enough to generate the PRD.
[PRD]
PRD: Dark Mode for Ralph TUI
Overview
Add dark mode support to ralph-tui to reduce eye strain during long orchestration sessions. The feature will apply to all TUI components and persist the user's preference in the config file.
Goals
- Provide a comfortable viewing experience in low-light environments
- Persist theme preference across sessions
- Apply consistent styling to all TUI components
Quality Gates
Epic-Level (run once on epic completion)
General codebase checks that run ONCE when all stories are done:
bun typecheck— type checking passes with zero errorsbun lint— linting passes with zero warnings
Story-Level (checked per story)
- UI stories: Verify in browser using dev-browser skill
User Stories
US-001: Add theme configuration [Schema]
Description: As a user, I want to set my preferred theme (light/dark) so that it persists across sessions.
Acceptance Criteria:
- Add
themefield to.ralph-tui.yamlschema - Support values: "light", "dark", "system"
- Default to "light" for backwards compatibility
- Verify field is read correctly by loading config in test
Mark each item [x] as you complete it. Only close when all are checked.
US-002: Create dark theme color palette [UI]
Description: As a user, I want a soft-contrast dark theme that's easy on the eyes.
Acceptance Criteria:
- Define dark palette with gray tones (not pure black) in theme constants file
- Ensure sufficient contrast ratios (WCAG AA) — check with contrast calculator
- Colors defined for all UI states: default, selected, hover, disabled
- Verify in browser using dev-browser skill
Mark each item [x] as you complete it. Only close when all are checked.
US-003: Apply theme to TUI components [UI]
Description: As a user, I want all TUI components to respect my theme preference.
Acceptance Criteria:
- Header component uses theme colors (verify in browser)
- Task list uses theme colors (verify in browser)
- Detail panels use theme colors (verify in browser)
- Progress bar uses theme colors (verify in browser)
- Dialogs use theme colors (verify in browser)
- Verify in browser using dev-browser skill
Mark each item [x] as you complete it. Only close when all are checked.
US-004: Add theme toggle in settings [UI]
Description: As a user, I want to toggle themes from within the TUI settings.
Acceptance Criteria:
- Theme option visible in settings view
- Changes apply immediately without restart (verify visually)
- Changes persist to config file (verify by reading
.ralph-tui.yamlafter toggle) - Verify in browser using dev-browser skill
Mark each item [x] as you complete it. Only close when all are checked.
Functional Requirements
- FR-1: Theme setting must be readable from
.ralph-tui.yaml - FR-2: Theme must apply on TUI startup
- FR-3: Theme changes in settings must apply immediately
- FR-4: All text must maintain readability in both themes
Non-Goals
- System theme auto-detection (future enhancement)
- Custom color schemes beyond light/dark
- Per-component theme overrides
Technical Considerations
- Use existing OpenTUI theming capabilities if available
- Consider creating a ThemeContext for React components
- Minimize re-renders when theme changes
Success Metrics
- All components render correctly in dark mode
- No accessibility contrast issues
- Theme persists across sessions
Open Questions
- Should we detect system theme preference automatically in v2? [/PRD]
Checklist
Before outputting the PRD:
- Asked clarifying questions with lettered options
- Asked about quality gates with two-tier classification (REQUIRED)
- Asked about testing trophy layers (REQUIRED)
- Asked follow-up questions when needed
- Quality Gates section has Epic-Level and Story-Level subsections
- Testing Trophy section exists with specific tests per layer (unit, contract, integration, frontend)
- Each backend story names its tests in acceptance criteria (TDD: test first)
- Each UI story has contract test for data shape from invoke
- Block-merge tests identified for implicit/background logic
- Epic-level gates are general commands (typecheck, lint, ci) — NOT in individual stories
- Story-level gates are assigned to relevant stories (UI → browser, backend → endpoint check)
- Stories tagged with type:
[Schema],[Backend],[UI],[Integration] - Every acceptance criterion is a verifiable
- [ ]checkbox - Every story includes "Mark each item [x]..." instruction
- User stories are small and independently completable
- Functional requirements are numbered and unambiguous
- Non-goals section defines clear boundaries
- PRD is wrapped in
[PRD]...[/PRD]markers