Plan

!../principles/SKILL_BODY.md

!../tool-gitnexus/SKILL_BODY.md

---

Sprint Planning Workflow

Core Principle

A plan is only as good as its dependency order. Tasks executed out of order produce integration failures that look like bugs but are actually sequencing errors. The goal is a dependency-ordered task list where each task has a clear skill assignment, a bounded scope, and a single testable acceptance condition.

Upstream skill: sextant:refine-requirements answers "what to build" and produces a requirements confirmation document. This skill answers "how to build it and in what order." If requirements are still unclear or unconfirmed, use sextant:refine-requirements first.

---

Complete Execution Workflow

Session Resume Check

Before Step 1, check whether .sextant/state.json exists in the project root.

If the file exists: display the current sprint summary (requirement name, task statuses) and ask:

"An existing sprint plan was found. Resume it, or start a new plan?"

• Resume: present the task list with current statuses; skip to the Execution Handoff section. Continue from the first pending or in_progress task.
• New plan: proceed with Steps 1–4 normally. Confirm with the user before overwriting the existing state.

If the file does not exist: proceed with Step 1 normally.

Step 1: Parse Inputs

Accept either a requirements confirmation document (output of sextant:refine-requirements) or a clearly stated, unambiguous requirement. Extract:

• Modules involved: which existing modules will be modified or extended?
• Entry points: where does the new behavior surface (API endpoint, CLI command, UI trigger, event producer)?
• Integration points: where does the new code connect to existing code (interfaces, events, shared data structures, registries)?
• Paradigm: which architecture paradigm applies (per §6.0)? — determines the default layer ordering in Step 2.

🔗 When GitNexus is available, use context / query MCP tools to extract module relationships automatically.

Step 2: Dependency Analysis

Ordering rule: lower-layer tasks must complete before higher-layer tasks that depend on them.

Default task ordering by paradigm (matches §6.2 dependency direction):

Paradigm	Default Task Order
Backend layered	Data layer → Logic layer → Entry layer → Tests
Frontend component tree	Store / hooks → Feature components → Page → Tests
CLI / Script	Core logic → Command handlers → CLI entry → Tests
Functional (FP)	Pure core functions → I/O adapters → Entry → Tests
Monorepo	Shared packages → Domain packages → App packages → Tests
Event-driven	Schema / event contracts → Producers → Consumers → Integration tests
Serverless	Core logic → Handler functions → Infrastructure config → Tests
AI/ML pipeline	Data pipeline → Preprocessing → Model → Evaluation → Serving

Cross-task data dependencies: if Task B consumes the output or interface produced by Task A, Task A must be sequenced first — regardless of layer.

Conflict arbitration: when ordering is ambiguous, apply §5.5 arbitration rules. When a task touches a public interface, treat it as at least Medium scale and confirm before sequencing.

Step 3: Task Specification

For each task, specify all six fields:

1. Title: one action verb + one noun (e.g., "Add UserRepository.findByEmail method")
2. Skill: which sextant sub-skill handles this task?
   • New code → sextant:add-feature
   • Change existing → sextant:modify-feature
   • Multi-module coordinated change → sextant:migrate
   • Tests → sextant:write-tests
   • Unclear requirements → sextant:refine-requirements
3. Scale: apply the Impact Radius Scorecard (§3.2) — Lightweight / Medium / Large
4. Files: likely files to create or modify
5. Depends on: list of task IDs that must complete first, or "None"
6. Acceptance: a single, testable condition confirming the task is done

Acceptance condition format: Given <precondition>, when <action>, then <verifiable outcome>. One sentence — if it requires more, the task scope is too large and should be split.

Scale gate: if any task scores as Architectural (Impact Radius 9–10), flag it for decomposition before proceeding. Do not plan an Architectural task as a single unit.

Step 4: Output the Plan

─── Sprint Plan ─────────────────────────────────────
Requirement: <name / one-sentence description>
Paradigm:    <detected paradigm>

Task 1: <title>
  Skill:       <sextant skill name>
  Scale:       <Lightweight / Medium / Large> (Impact radius: N)
  Files:       <likely files to create or modify>
  Depends on:  None / Task N, Task M
  Acceptance:  Given <precondition>, when <action>, then <verifiable outcome>.

Task 2: <title>
  Skill:       ...
  Scale:       ...
  Files:       ...
  Depends on:  ...
  Acceptance:  ...

Suggested sequence: 1 → 3 → 2 → 4 → 5
Rationale: <one sentence explaining any non-obvious ordering decisions>
─────────────────────────────────────────────────────

Execution Handoff

After the user confirms the sprint plan, this skill transitions from planner to pipeline entry point. Immediately offer to begin execution:

─── Ready to Execute ────────────────────────────────
The plan is confirmed. Suggested starting point:

Task 1: <title>
  → Invoke: sextant:<skill>
  → To begin, describe the task to me and I will apply the <skill> workflow.

When Task 1 is complete, Task 2 becomes unblocked:
  → Invoke: sextant:<skill>

Continue in sequence: <1 → 2 → ...>
─────────────────────────────────────────────────────

Flow rules for execution:

• Each task handoff is a fresh skill invocation — the executing skill owns its own workflow from that point
• If a task reveals a blocking issue (test failure, missing dependency, ambiguous requirement), pause execution and surface it before moving to the next task
• If a task's scope expands beyond its planned scale during execution, re-assess the Impact Radius before proceeding — it may affect downstream task ordering
• Tasks with no mutual dependencies can be executed in parallel if the user has multiple working contexts

State persistence (optional):

To resume this sprint across sessions, ask the user: "Save sprint state to .sextant/state.json so we can resume later?"

If the user confirms, write the file:

{
  "version": "1",
  "sprint": {
    "requirement": "<one-sentence requirement name>",
    "tasks": [
      {
        "id": 1,
        "title": "<title>",
        "skill": "sextant:<skill>",
        "scale": "Lightweight | Medium | Large",
        "status": "pending | in_progress | done | blocked",
        "depends_on": [],
        "acceptance": "<testable condition>",
        "flags": []
      }
    ],
    "suggested_sequence": [1, 2, 3]
  },
  "flags": []
}

Update the relevant task's status field as tasks complete:

• pending → in_progress when a task begins
• in_progress → done when its acceptance condition is met
• in_progress → blocked if a blocking issue is surfaced

Do not write .sextant/state.json without explicit user confirmation. Automatic writes are forbidden.

---

Forbidden Actions

• Do not assign vague acceptance criteria ("works correctly", "looks good") — every condition must be testable without interpretation
• Do not output an unordered list of tasks and call it a plan — dependency order is the core deliverable
• Do not plan a task that scores Architectural without first flagging it for decomposition
• Do not begin planning if the requirement still has unresolved 🔴 gaps — resolve them via sextant:refine-requirements first

---

Reply Format

Prepend the Sprint Plan block with a one-line summary:

Planning complete: <N> tasks, suggested sequence: <1 → 2 → ...>, scales: <N Lightweight, N Medium, N Large>.

Then output the full Sprint Plan block.