Imported from majiayu000/claude-skill-registry (skills/planning/ln-230-story-prioritizer/SKILL.md). Install upstream with npx skills add majiayu000/claude-skill-registry --skill ln-230-story-prioritizer. Copyright stays with the author (MIT).

Paths: File paths (references/, ../ln-*) are relative to this skill directory.

Story Prioritizer

Type: L3 Worker Category: 2XX Planning

Evaluate Stories using RICE scoring with market research. Generate consolidated prioritization table for Epic.

Purpose & Scope

• Prioritize Stories AFTER ln-220 creates them
• Triage all Stories cheaply before doing deep research
• Research market size and competition only where it changes prioritization confidence
• Calculate RICE score for each Story
• Generate prioritization table (P0/P1/P2/P3)
• Output: docs/market/[epic-slug]/prioritization.md

When to Use

Use this skill when:

• Stories created by ln-220, need business prioritization
• Planning sprint with limited capacity (which Stories first?)
• Stakeholder review requires data-driven priorities
• Evaluating feature ROI before implementation

Do NOT use when:

• Epic has no Stories yet (run ln-220 first)
• Stories are purely technical (infrastructure, refactoring)
• Prioritization already exists in docs/market/

Input Parameters

depth options:

• quick - 2-3 min/Story, 1 WebSearch per type
• standard - 5-7 min/Story, 2-3 WebSearches per type
• deep - 8-10 min/Story, comprehensive research

Output Structure

docs/market/[epic-slug]/
└── prioritization.md    # Consolidated table + RICE details + sources

Runtime Contract

MANDATORY READ: Load references/planning_worker_runtime_contract.md, references/coordinator_summary_contract.md MANDATORY READ: Load references/researchgraph_mcp_usage.md when Stories cite H/G/run IDs or project researchgraph evidence can change priority confidence.

Runtime family: planning-worker-runtime

Identifier:

• epic-{epicId}

Phases:

1. PHASE_0_CONFIG
2. PHASE_1_DISCOVERY
3. PHASE_2_LOAD_STORY_METADATA
4. PHASE_3_ANALYZE_STORIES
5. PHASE_4_GENERATE_PRIORITIZATION
6. PHASE_5_WRITE_SUMMARY
7. PHASE_6_SELF_CHECK

Summary contract:

• summary_kind=story-prioritization-worker
• payload includes epic_id, depth, stories_analyzed, priority_distribution, top_story_ids, prioritization_path, warnings
• managed mode writes to caller-provided summaryArtifactPath
• default managed artifact path pattern: .hex-skills/runtime-artifacts/runs/{parent_run_id}/story-prioritization-worker/ln-230--{identifier}.json

Table columns (from user requirements):

Inputs

Resolution: Epic Resolution Chain. Status filter: Active (planned/started)

Tools Config

MANDATORY READ: Load references/environment_state_contract.md, references/storage_mode_detection.md, references/input_resolution_pattern.md

Extract: task_provider = Task Management → Provider

Research Tools

Workflow

Phase 1: Discovery (2 min)

Objective: Validate input and prepare context.

Process:

1. Resolve epicId: Run Epic Resolution Chain per guide.

2. Load Epic details:

   • IF task_provider == "linear": get_project(query=epicId)
   • ELSE IF task_provider == "github": gh issue view {epicId} -R {REPO} --json number,title,body
   • ELSE: Read("docs/tasks/epics/epic-{N}-*/epic.md")
   • Extract: Epic ID, title, description

3. Auto-discover configuration:

   • Read docs/tasks/kanban_board.md for Team ID
   • Slugify Epic title for output path

4. Check existing prioritization:

   Glob: docs/market/[epic-slug]/prioritization.md
   

   • If exists: Ask "Update existing or create new?"
   • If new: Continue

5. Create output directory:

   mkdir -p docs/market/[epic-slug]/

Output: Epic metadata, output path, existing check result

Phase 2: Load Stories Metadata (3 min)

Objective: Build Story queue with metadata only and prepare rough scoring inputs for all Stories.

Process:

1. Query Stories from Epic: IF task_provider == "linear":

   list_issues(project=Epic.id, label="user-story")
   

   ELSE IF task_provider == "github":

   gh api /repos/{O}/{R}/issues/{epic_num}/sub_issues --jq '.[].number'
   → for each: gh issue view {num} -R {REPO} --json number,title,state,labels
   → filter: label "user-story"
   

   ELSE (file mode):

   Glob("docs/tasks/epics/epic-{N}-*/stories/*/story.md")
   

2. Extract metadata only:

   • Story ID, title, status
   • minimal Epic context if available
   • DO NOT load full descriptions yet

3. Filter Stories:

   • Exclude: Done, Cancelled, Archived
   • Include: Backlog, Todo, In Progress

4. Build processing queue:

   • Order by: existing priority (if any), then by ID
   • Count: N Stories to process

Output: Story queue (ID + title + minimal context), ~50-80 tokens/Story

Phase 3: Two-Pass Story Analysis

Objective: Score all Stories cheaply first, then spend deep research only on candidates where it changes the decision.

Critical: Keep maximum context to one full Story at a time even during deep research.

If a researchgraph layout exists, run local graph checks only for Stories whose priority depends on hypothesis status, goal coverage, benchmark evidence, or implementation readiness. Local graph evidence can change RICE confidence and risk; it does not replace market-size or competitor research.

Pass A: Cheap Triage For All Stories

For each Story, load only enough detail to estimate:

• customer problem
• rough solution shape
• likely reach
• likely impact
• likely effort
• initial confidence tier

Step 3.1: Load Story Description

IF task_provider == "linear":

// configured tracker provider: getStory(id=storyId)

ELSE IF task_provider == "github":

gh issue view {storyId} -R {REPO} --json number,title,body,state,labels

ELSE (file mode):

Read("docs/tasks/epics/epic-{N}-*/stories/us{NNN}-*/story.md")

Extract from Story:

• Feature: Story title
• Customer Problem: From "So that [value]" + Context section
• Solution: From Technical Notes (implementation approach)
• Rationale: From AC + Success Criteria

Step 3.2: Build rough RICE estimate

Use Story + Epic context to assign:

• rough Reach
• rough Impact
• rough Effort
• initial Confidence

Mark one of:

• full_research_required
• rough_estimate_ok
• borderline_needs_review

Send to Pass B only if:

• candidate looks P0/P1 on rough score
• confidence is low
• Story is near a priority threshold
• Story has strategic or market-sensitive uncertainty

Pass B: Selective Deep Research

Only for Stories selected in Pass A, run full external research.

Step 3.3: Research Market Size

WebSearch queries (based on depth):

"[customer problem domain] market size TAM {current_year}"
"[feature type] industry market forecast"

mcp__Ref query:

"[domain] market analysis Gartner Statista"

Extract:

• Market size: $XB (with unit: B=Billion, M=Million)
• Growth rate: X% CAGR
• Sources: URL + date

Confidence mapping:

• Industry report (Gartner, Statista) → Confidence 0.9-1.0
• News article → Confidence 0.7-0.8
• Blog/Forum → Confidence 0.5-0.6

Step 3.4: Research Competition

WebSearch queries:

"[feature] competitors alternatives {current_year}"
"[solution approach] market leaders"

Count competitors and classify:

Step 3.5: Calculate final RICE Score

RICE = (Reach x Impact x Confidence) / Effort

Reach (1-10): Users affected per quarter

Impact (0.25-3.0): Business value

Confidence (0.5-1.0): Data quality (from Step 3.2)

Data Confidence Assessment:

For each RICE factor, assess data confidence level:

Output: Show confidence per factor in prioritization table + RICE range (optimistic/pessimistic) to make uncertainty explicit.

Effort (1-10): Person-months

Step 3.6: Determine Priority

Step 3.7: Store and Clear

• Append row to in-memory results table
• Mark whether row is full-research or rough-estimate
• Clear Story description from context
• Move to next Story in queue

Output per Story: Complete row for prioritization table with confidence tier

Phase 4: Generate Prioritization Table (5 min)

Objective: Create consolidated markdown output.

Process:

1. Sort results:

   • Primary: Priority (P0 → P3)
   • Secondary: RICE score (descending)

2. Generate markdown:

   • Use template from references/templates/prioritization_template.md
   • Fill: Priority Summary, Main Table, RICE Details, Sources
   • Explicitly show whether each Story used full research or rough estimate

3. Save file:

   Write: docs/market/[epic-slug]/prioritization.md
   

Output: Saved prioritization.md

Phase 5: Summary & Next Steps (1 min)

Objective: Display results and recommendations.

Output format:

## Prioritization Complete

**Epic:** [Epic N - Name]
**Stories analyzed:** X
**Time elapsed:** Y minutes

### Priority Distribution:
- P0 (Critical): X Stories - Implement ASAP
- P1 (High): X Stories - Next sprint
- P2 (Medium): X Stories - Backlog
- P3 (Low): X Stories - Consider deferring

### Top 3 Priorities:
1. [Story Title] - RICE: X, Market: $XB, Competition: Blue/Red

### Saved to:
docs/market/[epic-slug]/prioritization.md

### Next Steps:
1. Review table with stakeholders
2. Run ln-300 for P0/P1 Stories first
3. Consider cutting P3 Stories

Time-Box Constraints

Time management rules:

• If Story exceeds time budget: keep rough estimate, mark lower confidence
• If total exceeds budget: reserve deep research only for high-potential or borderline Stories
• Parallel WebSearch where possible (market + competition)

Token Efficiency

Loading pattern:

• Phase 2: Metadata only (~50 tokens/Story)
• Phase 3: Full description ONE BY ONE (~3,000-5,000 tokens/Story)
• After each Story: Clear description, keep only result row (~100 tokens)

Memory management:

• Sequential processing (not parallel)
• Maximum context: 1 Story description at a time
• Results accumulate as compact table rows

Integration with Ecosystem

Position in workflow:

ln-210 (Scope → Epics)
     ↓
ln-220 (Epic → Stories)
     ↓
ln-230 (RICE per Story → prioritization table) ← THIS SKILL
     ↓
ln-300 (Story → Tasks)

Dependencies:

• WebSearch, mcp__Ref (market research)
• Task provider: configured tracker provider (load Epic, Stories) per references/tracker_provider_contract.md
• Glob, Write, Bash (file operations)

Downstream usage:

• Sprint planning uses P0/P1 to select Stories
• ln-300 processes Stories in priority order
• Stakeholders review before implementation

Structured worker output:

• return the prioritization summary envelope even in standalone mode
• write the same JSON artifact when summaryArtifactPath is provided

Critical Rules

1. Triage first - do cheap scoring across all Stories before deep research
2. Source all deep-research data - every Market number needs source + date
3. Prefer recent data - last 2 years, warn if older
4. Cross-reference when depth justifies it - use 2+ sources for market-sensitive Stories
5. Time-box strictly - keep rough estimates when deeper research will not change the decision
6. Confidence levels - mark High/Medium/Low and whether score is rough or full-research
7. No speculation - only sourced claims, note "[No data]" gaps
8. One Story at a time - token efficiency critical
9. Preserve language - if user asks in Russian, respond in Russian

Definition of Done

• Epic validated via the configured tracker provider
• All Stories loaded through metadata-first queue
• Pass A rough triage completed for all Stories
• Deep research limited to high-potential or low-confidence Stories
• RICE score calculated for each Story
• Competition index assigned (1-5)
• Priority assigned (P0/P1/P2/P3)
• Confidence tier and research depth visible in output
• Table sorted by Priority + RICE
• File saved to docs/market/[epic-slug]/prioritization.md
• Summary with top priorities and next steps
• Structured story-prioritization-worker summary returned
• Summary artifact written when summaryArtifactPath is provided
• Total time within budget

Example Usage

Basic usage:

ln-230-story-prioritizer epic="Epic 7"

With parameters:

ln-230-story-prioritizer epic="Epic 7: Translation API" depth="deep"

Specific Stories:

ln-230-story-prioritizer epic="Epic 7" stories="US001,US002,US003"

Example output (docs/market/translation-api/prioritization.md):

Phase 6: Meta-Analysis

Optional reference: load references/meta_analysis_protocol.md only when the user asks for post-run meta-analysis or protocol-formatted run reflection.

Skill type: planning-worker. When requested, run after all phases complete. Output to chat using the planning-worker format.

Reference Files

• MANDATORY READ: Load references/environment_state_contract.md
• MANDATORY READ: Load references/storage_mode_detection.md
• MANDATORY READ: Load references/research_tool_fallback.md

Version: 2.0.0 Last Updated: 2026-04-05