Writing Plans

Writing Plans

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.

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

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

Plan Document Header

Every plan MUST start with this header:

# [Feature Name] Implementation Plan

> **For Claude:** REQUIRED SUB-SKILL: Use the executing-plans skill 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**

```python
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**

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

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

Common Pitfalls

1. Plans Too Abstract (No Concrete Paths or Commands)

Symptom: Plan reads smoothly but lacks actionable details. Steps like "Add validation" or "Update the API handler" leave the engineer guessing which file, which function, or what commands to run.

Why it fails:

• Engineer wastes time hunting for the right file or inferring code structure
• Different engineers interpret abstract steps differently, leading to rework
• No way to verify the plan is complete until someone starts implementing

How to fix:

• Always include exact file paths: src/handlers/user.ts not update the handler
• Always include exact commands: pytest tests/handlers/test_user.py -v not run tests
• Always include complete code in fenced blocks, not pseudocode
• Provide line numbers when modifying existing files: Modify: src/handlers/user.ts:45-62

Example (bad vs good):

❌ BAD: "Update the validation logic to handle email addresses"
✅ GOOD: "Modify: src/validators/email.py:12-18. Replace the regex pattern with: re.compile(r'^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}$')"

2. Task Granularity Wrong (Steps Too Large or Too Small)

Symptom: Steps are either vague epics ("Build the authentication system") or so granular they create friction ("Import the module", "Define the function signature", "Write the function body").

Why it fails:

• Large steps: engineer lacks guardrails; easy to implement wrong thing
• Small steps: excessive context switching; commits become noise; hard to trace intent
• Either way, progress becomes hard to measure and verify

How to fix:

• Target 2-5 minute steps: small enough to verify and commit independently, large enough to be meaningful
• Group related micro-steps under a single task heading, then break them into numbered steps
• Each step should produce a git-committable change
• Use this litmus test: "Can I understand and verify this single commit in isolation?"

Example (bad vs good):

❌ BAD:
  Step 1: Import json
  Step 2: Define function
  Step 3: Write body
  Step 4: Run test

✅ GOOD:
  Step 1: Write failing test for json parsing
  Step 2: Implement minimal function that passes test
  Step 3: Run test and verify PASS
  Step 4: Commit

3. Missing Verification Steps (No Way to Confirm Task Success)

Symptom: Plan describes what to do but never says how to verify it's done correctly. Engineer finishes a step and isn't sure if it actually worked.

Why it fails:

• Silent failures: code runs but produces wrong result; engineer doesn't catch it
• Wasted debugging time: engineer has to invent their own verification method
• Plan can't be reused reliably because success criteria are ambiguous

How to fix:

• Every step that changes behavior must include a verification command
• Verification command must include expected output or success indicator
• For tests: Expected: PASS. For commands: Expected output: [actual output]
• Include negative cases: "Run: pytest tests/... -v. Expected: Should fail with 'ValidationError'"

Example (bad vs good):

❌ BAD:
  Step 3: Implement the validation function

✅ GOOD:
  Step 3: Implement the validation function
  
  ```python
  def validate_email(email):
      pattern = r'^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}$'
      return bool(re.match(pattern, email))
  ```

  Step 4: Run test to verify it passes
  
  Run: `pytest tests/validators/test_email.py::test_valid_email -v`
  Expected: `PASSED test_valid_email`

4. DRY Violations (Repeated Instructions Across Tasks)

Symptom: Plan repeats the same instruction in multiple tasks ("Run the full test suite", "Import X module", "Follow the naming convention"). Same boilerplate appears in steps across different tasks.

Why it fails:

• Maintainability: updating the pattern requires editing multiple places
• Cognitive load: engineer has to understand the intent is the same across tasks
• Error-prone: same instruction might be subtly different in different places, causing confusion

How to fix:

• Capture repeated patterns in a "Setup" or "Prerequisites" section at the top of the plan
• Use consistent command syntax and structure across similar steps
• Link to shared instructions: "As per 'Testing' section above, run: pytest ..."
• If a pattern is truly repeated, ask: can these tasks be combined or is there a shared utility that would eliminate the duplication?

Example (bad vs good):

❌ BAD:
  Task 1: Write test, run `pytest tests/handlers/test_a.py -v`, commit
  Task 2: Write test, run `pytest tests/handlers/test_b.py -v`, commit
  Task 3: Write test, run `pytest tests/handlers/test_c.py -v`, commit

✅ GOOD:
  # Testing Pattern
  After each implementation, verify with: `pytest tests/handlers/test_<name>.py -v`
  Expected: All tests PASS
  
  Task 1: [test_a implementation]
  Task 2: [test_b implementation]
  Task 3: [test_c implementation]

5. Assumes Context Agent Won't Have (Framework Specifics, Conventions)

Symptom: Plan references "the standard pattern", "as we do it", or "use the company convention" without explaining what that actually means. Assumes engineer knows framework shortcuts, naming conventions, or project structure.

Why it fails:

• Fresh engineer or external contributor has no reference for "the pattern"
• Different projects have different conventions; agent may mix them up
• Engineer spends time reverse-engineering conventions instead of following a plan

How to fix:

• Explicitly state conventions in a "Context" or "Setup" section at the plan start
• Name the convention if it's framework-specific: "Django model naming: class MyModel(models.Model)"
• Provide an example: "All API handlers follow this pattern: [example file + code snippet]"
• If a pattern is specific to this codebase, explain it briefly instead of assuming knowledge

Example (bad vs good):

❌ BAD:
  Step 1: Create a handler using the standard pattern
  Step 2: Follow our naming convention for tests

✅ GOOD:
  Context: This project uses Flask. All handlers live in src/handlers/ and follow the pattern:
  ```python
  @app.route('/endpoint', methods=['POST'])
  def handle_endpoint():
      # Handler logic
      return jsonify({...})
  ```
  
  Step 1: Create handler in src/handlers/my_handler.py following the pattern above
  Step 2: Create test in tests/handlers/test_my_handler.py. Naming: test_<function_name>_<scenario>

Execution Handoff

After saving the plan, offer execution choice:

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

1. Subagent-Driven (this session) - I dispatch fresh subagent per task, review between tasks, fast iteration

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

Which approach?"

If Subagent-Driven chosen:

• Spawn a subagent per task, review results between tasks, fast iteration
• Stay in this session and orchestrate

If Parallel Session chosen:

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