shipyard-writing-plans

Writing Plans

When This Skill Activates

• You have a spec, requirements, or design for a multi-step implementation task
• You need to break work into bite-sized, executable tasks before touching code
• You are preparing work for builder agents or a parallel execution session

Announce at start: "I'm using the writing-plans skill to create the implementation plan."

Context: This should be run in a dedicated worktree (created by brainstorming skill).

Save plans to: docs/plans/YYYY-MM-DD-<feature-name>.md

Natural Language Triggers

• "write a plan", "create a plan", "plan this feature", "break this down into tasks"

Overview

Write comprehensive implementation plans assuming the engineer has zero context for our codebase and questionable taste. Document everything they need to know: which files to touch for each task, code, testing, docs they might need to check, how to test it. Give them the whole plan as bite-sized tasks. DRY. YAGNI. TDD. Frequent commits.

Assume they are a skilled developer, but know almost nothing about our toolset or problem domain. Assume they don't know good test design very well.

Shipyard Plan Format

Shipyard plans use XML-structured tasks with verification criteria. Each task includes:

<task id="1" name="Component Name">
  <description>What this task accomplishes</description>
  <files>
    <create>exact/path/to/file.py</create>
    <modify>exact/path/to/existing.py:123-145</modify>
    <test>tests/exact/path/to/test.py</test>
  </files>
  <steps>
    <step>Write the failing test</step>
    <step>Run test to verify it fails</step>
    <step>Write minimal implementation</step>
    <step>Run test to verify it passes</step>
    <step>Commit</step>
  </steps>
  <verification>
    <command>pytest tests/path/test.py::test_name -v</command>
    <expected>PASS</expected>
  </verification>
</task>

This structured format enables /shipyard:build to parse and execute tasks systematically, and /shipyard:status to track progress.

Task Granularity Guide

Size	Example	Action
Too big	"Implement authentication system"	Split — no single commit for a whole system
Right size	"Add JWT token validation middleware"	Keep — one TDD cycle, one commit
Too small	"Add import statement"	Merge with its parent task

Target: Each task = one TDD cycle (write test → fail → implement → pass → commit). If a task needs more than one commit, split it.

Bite-Sized Task Granularity

Each step is one action (2-5 minutes):

• "Write the failing test" - step
• "Run it to make sure it fails" - step
• "Implement the minimal code to make the test pass" - step
• "Run the tests and make sure they pass" - step
• "Commit" - step

Coupling Detection

Before ordering tasks, check for dependencies. Tasks that must share state or touch the same file need sequencing:

Dependency Type	Example	Resolution
Same file	Tasks A and B both modify auth.py	Sequence them; never parallelize
Import dependency	Task B imports what Task A creates	B blocks on A
Interface contract	Task B depends on Task A's return type	Define interface in Task A, implement in B
Shared utility	Both tasks call a helper that doesn't exist yet	Create helper as Task 0

Red flag: Two tasks listed as parallelizable that both modify the same file — this will produce merge conflicts.

Mid-Phase Adaptation

When reality diverges from the plan during execution:

• Minor divergence (file path changed, one extra step) — adapt in place, note in SUMMARY.md
• Major divergence (wrong architecture, missing component) — pause, update plan, re-approve before continuing
• Blocker (dependency missing, API changed) — stop, document in SUMMARY.md as blocker, escalate

Do not silently adapt major changes. The plan is a contract; changes need acknowledgment.

Plan Document Header

Every plan MUST start with this header:

# [Feature Name] Implementation Plan

> **For Claude:** REQUIRED SUB-SKILL: Use shipyard:shipyard-executing-plans to implement this plan task-by-task.

**Goal:** [One sentence describing what this builds]

**Architecture:** [2-3 sentences about approach]

**Tech Stack:** [Key technologies/libraries]

---

Task Structure

### Task N: [Component Name]

**Files:**
- Create: `exact/path/to/file.py`
- Modify: `exact/path/to/existing.py:123-145`
- Test: `tests/exact/path/to/test.py`

**Step 1: Write the failing test**

```python
def test_specific_behavior():
    result = function(input)
    assert result == expected

Step 2: Run test to verify it fails

Run: pytest tests/path/test.py::test_name -v Expected: FAIL with "function not defined"

Step 3: Write minimal implementation

def function(input):
    return expected

Step 4: Run test to verify it passes

Run: pytest tests/path/test.py::test_name -v Expected: PASS

Step 5: Commit

git add tests/path/test.py src/path/file.py
git commit -m "feat: add specific feature"

</instructions>

<examples>

## Example: Well-Written vs Poorly-Written Plan Task

<example type="good" title="Clear, executable task with exact paths and code">
### Task 2: Add Email Validation

**Files:**
- Create: `src/validators/email.py`
- Test: `tests/validators/test_email.py`

**Step 1: Write the failing test**
```python
def test_rejects_empty_email():
    with pytest.raises(ValidationError, match="email is required"):
        validate_email("")

Step 2: Run test to verify it fails Run: pytest tests/validators/test_email.py::test_rejects_empty_email -v Expected: FAIL with "cannot import name 'validate_email'"

Step 3: Write minimal implementation

from src.errors import ValidationError

def validate_email(email: str) -> str:
    if not email or not email.strip():
        raise ValidationError("email is required")
    return email.strip()

Step 4: Run test to verify it passes Run: pytest tests/validators/test_email.py::test_rejects_empty_email -v Expected: PASS

Step 5: Commit

git add src/validators/email.py tests/validators/test_email.py
git commit -m "feat: add email validation with empty check"

### Task 2: Add Validation

Add email validation to the project. Make sure it handles edge cases. Write tests.

The good example provides exact file paths, complete code, exact commands with expected output, and a single commit per TDD cycle. The bad example forces the builder to guess paths, invent code, and figure out what "edge cases" means.

Remember

• Exact file paths always
• Complete code in plan (not "add validation")
• Exact commands with expected output
• Reference relevant skills with @ syntax
• DRY, YAGNI, TDD, frequent commits

Red Flags

A plan is not ready to ship if any task:

• Has no file path (builder must guess where to put the code)
• Has no verification command (builder cannot confirm it worked)
• Touches the same file as an adjacent task (guaranteed conflict)
• Would take more than 20 minutes (too big — split it)
• Mentions "implement feature X" without TDD steps (builder will skip tests)

AI-Awareness: Common Plan Quality Failures

AI writers produce plans with predictable failure modes:

Failure	Symptom	Fix
Tightly-coupled tasks	Two tasks listed as parallel both modify same file	Add dependency, sequence them
Vague verification	<expected>success</expected>	Specify exact output or exit code
Missing file paths	"Add auth middleware" without path	Read codebase, provide exact path
Over-scoped tasks	Single task covers auth + session + logging	Split into one concern per task
Missing TDD steps	Steps go straight to implementation	Add write-test, run-test, implement, verify

Rationalization Prevention

Excuse	Reality
"The path is obvious"	Builders hallucinate paths when not given them. Provide exact paths.
"I'll flesh out the test later"	"Later" = never in execution pressure. Write test code in the plan.
"These tasks are independent enough"	If they touch the same file, they are coupled. No exceptions.
"The verification step is implied"	Implied verification is skipped verification. Write the command.
"The task is simple, no need to split"	Simple tasks that take >20 min are not simple. Split them.

Execution Handoff

After saving the plan, offer execution choice:

"Plan complete and saved to docs/plans/<filename>.md. Two execution options:

1. Agent-Driven (this session) - I dispatch fresh builder agent per task, review between tasks, fast iteration

2. Parallel Session (separate) - Open new session with executing-plans, batch execution with checkpoints

Which approach?"

If Agent-Driven chosen:

• REQUIRED SUB-SKILL: Use shipyard:shipyard-executing-plans
• Stay in this session
• Fresh builder agent per task + two-stage review (spec compliance then code quality)

If Parallel Session chosen:

• Guide them to open new session in worktree
• REQUIRED SUB-SKILL: New session uses shipyard:shipyard-executing-plans

Integration

Called by: shipyard:shipyard-brainstorming — after requirements are gathered, planning begins Pairs with: shipyard:shipyard-tdd — TDD steps are embedded in every task Leads to: shipyard:shipyard-executing-plans — plan is handed off for execution