mcpField
L
ln-220-story-coordinator

ln-220-story-coordinator

github.com/levnikolaevichDevelopment584375 installsUpdated 2d ago
Repository

About ln-220-story-coordinator

A runtime-backed skill for coordinating Story planning from an Epic. It assembles context, builds an ideal plan before inspecting existing Stories, then routes work to creation or replanning workers.

Key features

  1. 01Assembles Epic planning context
  2. 02Researches constraints affecting technical notes
  3. 03Builds an ideal Story plan first
  4. 04Routes Stories and detects planning mode
  5. 05Delegates to standalone workers
  6. 06Tracks runtime phases and worker results

Use cases

  1. 01Breaking an Epic into planned Stories
  2. 02Replanning or adding Stories for an Epic
  3. 03Coordinating Story planning with pause and resume

MCP support

When researchgraph references or layout affect Story decomposition, the skill uses its MCP workflow to verify the index and query relevant goals, hypotheses, evidence, or runs.

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/levnikolaevich/claude-code-skills --skill ln-220-story-coordinator

After installing, type/ln-220-story-coordinatorin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

references/
agent_skill_roots_contract.md
coordinator_runtime_contract.md
coordinator_summary_contract.md
environment_state_contract.md
meta_analysis_protocol.md
problem_solving.md
replan_algorithm.md
researchgraph_mcp_usage.md
storage_mode_detection.md
references/scripts/coordinator-runtime/lib/
artifacts.mjs
cli-helpers.mjs
core.mjs
loop-health.mjs
runtime-constants.mjs
schemas.mjs
validate.mjs
worker-guards.mjs
worker-runtime.mjs
references/scripts/planning-runtime/lib/
cli.mjs
guards.mjs
store.mjs
template-compliance.mjs
references/scripts/planning-worker-runtime/
cli.mjs
references/scripts/planning-worker-runtime/lib/
guards.mjs
phases.mjs
store.mjs
references/scripts/story-planning-runtime/
cli.mjs
references/scripts/story-planning-runtime/lib/
guards.mjs
phases.mjs
store.mjs

Details

From the project README

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

Story Coordinator

Type: L2 Domain Coordinator Category: 2XX Planning

Runtime-backed Story planning coordinator. The runtime owns flow control, pause/resume, and worker result tracking.

MANDATORY READ

Load these before execution:

  • references/coordinator_runtime_contract.md
  • references/story_planning_runtime_contract.md
  • references/coordinator_summary_contract.md
  • references/environment_state_contract.md
  • references/storage_mode_detection.md
  • references/problem_solving.md

MANDATORY READ: Load references/researchgraph_mcp_usage.md when the Epic, source docs, or existing Stories cite H/G/run IDs or when a researchgraph layout can change Story decomposition.

Purpose

  • assemble Epic planning context
  • perform focused standards research when it changes Technical Notes
  • build the ideal Story plan before checking existing Stories
  • detect routing and mode per epic group
  • batch child manifest/artifact preparation before delegation
  • delegate creation or replanning to standalone workers

Inputs

ParameterRequiredDescription
epicIdYesEpic to decompose
autoApproveNoIf false, runtime pauses on preview confirmation

Runtime

Runtime family: story-planning-runtime

Identifier:

  • epic-{epicId}

Phases:

  1. PHASE_0_CONFIG
  2. PHASE_1_CONTEXT_ASSEMBLY
  3. PHASE_2_RESEARCH
  4. PHASE_3_PLAN
  5. PHASE_4_ROUTING
  6. PHASE_5_MODE_DETECTION
  7. PHASE_6_DELEGATE
  8. PHASE_7_FINALIZE
  9. PHASE_8_SELF_CHECK

Terminal phases:

  • DONE
  • PAUSED

Summary flow:

  • consume child story-plan-worker artifacts from ln-221 / ln-222
  • write coordinator story-plan artifact during PHASE_7_FINALIZE

Phase Map

Phase 1: Context Assembly

Resolve Epic and assemble only the planning inputs that change Story decomposition:

  • Epic scope
  • success criteria
  • known personas and constraints
  • project task provider

Checkpoint payload:

  • context_ready
Phase 2: Research

Do focused research only when it changes Story Technical Notes or implementation constraints.

Do not let research expand Story scope.

If a researchgraph layout exists, use verify_index first and then query only the relevant goals, hypotheses, evidence, or runs. Use graph results to ground Story scope, readiness, and Technical Notes; do not create new Story scope just because graph debt exists.

Checkpoint payload:

  • research_status
  • research_file
Phase 3: Plan

Build the ideal Story plan before looking at existing Stories.

Rules:

  • vertical slices only
  • 5-10 Stories when Epic warrants it
  • concise Story statements and observable ACs
  • no orchestration prose inside the plan
  • delegate ACs must specify what equips the actor (context, instructions, tools, configuration), not just what the actor does

Checkpoint payload:

  • ideal_plan_summary
Phase 4: Routing

Route planned Stories to epic groups.

Fast path:

  • all Stories stay in the resolved Epic

Pause only when routing is ambiguous or requires confirmation.

Checkpoint payload:

  • routing_summary
Phase 5: Mode Detection

Determine mode per epic group:

  • CREATE
  • REPLAN
  • ADD

Checkpoint payload:

  • epic_group_modes
Phase 6: Delegate

Phase 6 has two internal steps.

Phase 6a: Prepare delegation

  • finalize routing groups
  • materialize worker manifests
  • precompute run_id and summary_artifact_path for each child
  • checkpoint the expected worker set before execution

Phase 6b: Execute delegation

Delegate by group:

  • ln-221-story-creator
  • ln-222-story-replanner

Workers remain standalone-capable. In managed mode the coordinator starts them through planning-worker-runtime, passes runId + summaryArtifactPath, stores the launch metadata in child_run, then records the resulting worker artifact through record-epic.

Worker summary kind:

  • story-plan-worker
Phase 7: Finalize

Finalize only after all expected worker summaries are recorded.

Coordinator output:

  • build one story-plan summary for the parent runtime
  • write it through node references/scripts/story-planning-runtime/cli.mjs record-plan-summary
  • persist the artifact before advancing to PHASE_8_SELF_CHECK

Template compliance gate: Fetch each created Story via the configured tracker provider (getStory operation per references/tracker_provider_contract.md). Run validateTemplateCompliance(description, 'story') from planning-runtime/lib/template-compliance.mjs. All stories must pass (9 sections in order). Record template_compliance_passed in state. Guard blocks SELF_CHECK without it.

Checkpoint payload:

  • final_result
  • template_compliance_passed
Phase 8: Self-Check

Confirm:

  • phase coverage
  • planned vs produced Story counts
  • no missing epic groups

Checkpoint payload:

  • pass
  • final_result

Pending Decisions

Use runtime PAUSED + pending_decision for:

  • missing context
  • routing confirmation
  • ambiguous ADD vs REPLAN
  • preview confirmation when autoApprove=false

Do not hold these decisions only in chat.

Worker Contract

Workers:

  • do not know the coordinator
  • do not read runtime state
  • remain standalone
  • may receive summaryArtifactPath
  • return shared summary envelope either way

Expected summary kind:

  • child workers: story-plan-worker
  • coordinator output: story-plan

Worker Invocation (MANDATORY)

Host Skill Invocation: Skill(skill: "...", args: "...") is mandatory delegation.

  • Claude: call the Skill tool exactly as shown.
  • Codex: if no Skill tool exists, locate the named skill in available skills, read its SKILL.md, treat args as $ARGUMENTS, execute that skill workflow, then return here with its result/artifact.
  • Do not inline worker logic or mark the worker complete without executing the target skill.
PhaseWorkerContext
6ln-221-story-creatorCREATE or ADD path
6ln-222-story-replannerREPLAN path
node references/scripts/planning-worker-runtime/cli.mjs start --skill {worker} --identifier {identifier} --manifest-file {workerManifestPath} --run-id {childRunId} --summary-artifact-path {childSummaryArtifactPath}
child_run = { skill, run_id, identifier, summary_artifact_path }
childSummaryArtifactPath = .hex-skills/runtime-artifacts/runs/{parent_run_id}/story-plan-worker/{worker}--{identifier}.json
Skill(skill: "{worker}", args: "{identifier} --ideal-plan {idealPlanJSON} --epic {epicId} --run-id {childRunId} --summary-artifact-path {childSummaryArtifactPath}")
Read {childSummaryArtifactPath}
node references/scripts/story-planning-runtime/cli.mjs record-epic --epic {epicId} --payload-file {childSummaryArtifactPath}
node references/scripts/story-planning-runtime/cli.mjs record-plan-summary --epic {epicId} --payload-file {coordinatorSummaryPath}

TodoWrite format (mandatory)

- Phase 1: Assemble context (pending)
- Phase 2: Research only what changes Technical Notes (pending)
- Phase 3: Build ideal Story plan (pending)
- Phase 4: Route Stories by Epic (pending)
- Phase 5: Detect mode per group (pending)
- Phase 6a: Prepare delegation batch (pending)
- Phase 6b: Execute worker(s) sequentially (pending)
- Phase 7: Finalize result (pending)
- Phase 8: Self-check (pending)

Critical Rules

  • Build the ideal plan before checking existing Stories.
  • Use research only to improve Technical Notes and implementation realism.
  • Do not keep routing or preview approvals in chat-only state.
  • Batch only read-only preparation. Do not parallelize Story mutations across routed groups unless runtime semantics explicitly allow it.
  • Do not create or update Stories directly when a worker should do it.
  • Consume worker summaries, not free-text worker prose.

Definition of Done

  • Runtime started with Epic-scoped identifier
  • Context assembly checkpointed
  • Research checkpointed or explicitly minimal
  • Ideal plan checkpointed
  • Routing checkpointed
  • Mode detection checkpointed
  • All expected worker summaries recorded
  • Coordinator story-plan summary recorded
  • Final result checkpointed
  • Template compliance passed for all created Stories
  • Self-check passed

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-coordinator. When requested, run after all phases complete. Output to chat using the protocol format.

Reference Files

  • references/environment_state_contract.md
  • references/storage_mode_detection.md
  • references/replan_algorithm.md

Version: 5.0.0 Last Updated: 2026-02-03

Comments

Markdown supported