Writing Implementation Plans

PURPOSE: Turn an approved design into a comprehensive implementation plan that an engineer (or AI agent) with zero codebase context can follow task by task. Every task is bite-sized, TDD-oriented, and self-contained.

CONFIGURATION: Reads jig.config.md for commit conventions, execution strategy preferences, and parallel threshold.

---

When to Use

Invoke this skill when:

• You have an approved design from brainstorm
• You have requirements (from a PRD, ticket, or conversation) for a multi-step task
• kickoff routes here during the PLAN stage
• The user says "write a plan", "create an implementation plan", or "/plan"

Do NOT use when:

• You do not have an approved design or clear requirements (use brainstorm first)
• The task is a single atomic change (just do it)

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

---

Scope Check

If the spec covers multiple independent subsystems, it should have been broken into sub-project specs during brainstorming. If it was not, suggest breaking this into separate plans -- one per subsystem. Each plan should produce working, testable software on its own.

---

Plan Document Header

Every plan MUST start with this header:

# [Feature Name] Implementation Plan

> **PRD:** docs/plans/YYYY-MM-DD-<topic>-prd.md *(include if a PRD exists)*
> **Design:** docs/plans/YYYY-MM-DD-<topic>-design.md *(include if a design doc exists)*
> **For agents:** Use team-dev (parallel) or sdd (sequential) to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.

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

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

**Tech Stack:** [Key technologies/libraries relevant to this plan]

---

The > **PRD:** and > **Design:** lines are how downstream spec reviewers find the acceptance checklist and design decisions. Always include them when those documents exist.

---

File Structure

Before defining tasks, map out which files will be created or modified and what each one is responsible for. This is where decomposition decisions get locked in.

## File Structure

| File | Action | Responsibility |
|------|--------|---------------|
| `exact/path/to/file.ext` | Create | Brief description |
| `exact/path/to/existing.ext` | Modify | What changes and why |
| `tests/exact/path/to/test.ext` | Create | What it tests |

File Structure Guidelines

• Design units with clear boundaries and well-defined interfaces. Each file should have one clear responsibility.
• AI agents reason best about code they can hold in context at once. Prefer smaller, focused files over large ones that do too much.
• Files that change together should live together. Split by responsibility, not by technical layer.
• In existing codebases, follow established patterns. If the codebase uses large files, do not unilaterally restructure -- but if a file you are modifying has grown unwieldy, including a split in the plan is reasonable.

---

Bite-Sized Task Granularity

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

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

Tasks should be scoped so an engineer can complete one in a focused burst without needing to context-switch. If a task requires more than 5 minutes of active work, break it down further.

---

Task Structure

Every task follows this template:

### Task N: [Component Name]

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

**Dependencies:** Requires Task M (if applicable)

- [ ] **Step 1: Write the failing test**

```language
// Test code here -- complete, runnable, no placeholders
```

- [ ] **Step 2: Run test to verify it fails**

Run: `<exact test command>`
Expected: FAIL with "<expected failure message>"

- [ ] **Step 3: Write minimal implementation**

```language
// Implementation code here -- complete, no placeholders
```

- [ ] **Step 4: Run test to verify it passes**

Run: `<exact test command>`
Expected: PASS

- [ ] **Step 5: Commit**

```bash
git add <specific files>
git commit -m "<message following project commit convention>"
```

Task Structure Requirements

• Exact file paths -- always. No "create a file in the appropriate directory."
• Complete code -- every step that changes code shows the complete code block. No summaries.
• Exact commands -- with expected output. The engineer should know what success looks like.
• Dependencies declared -- if Task N requires Task M, say so with blockedBy or Dependencies.
• Commit messages -- follow the project's commit convention from jig.config.md.

---

No Placeholders

Every step must contain the actual content an engineer needs. These are plan failures -- never write them:

Placeholder	Why It Fails
"TBD", "TODO", "implement later"	Engineer stops dead, has to figure it out
"Add appropriate error handling"	What errors? What handling? Be specific.
"Add validation"	What validation? For what inputs?
"Handle edge cases"	Which edge cases? List them.
"Write tests for the above"	Without actual test code? Useless.
"Similar to Task N"	The engineer may read tasks out of order. Repeat the code.
Steps describing what to do without showing how	Code steps require code blocks.
References to types/functions not defined in any task	Undefined = broken.

---

TDD Orientation

Plans are TDD-oriented by default:

1. Every feature task starts with a failing test
2. The test is run and confirmed failing
3. Minimal implementation makes the test pass
4. Tests are confirmed passing
5. Then commit

REQUIRED: Reference tdd skill for implementers. Subagents and teammates executing this plan should follow the TDD red-green-refactor cycle.

For tasks that are purely structural (creating directories, config files, boilerplate with no logic), TDD steps can be simplified to "create file, verify it exists, commit."

---

Self-Review

After writing the complete plan, review it with fresh eyes. This is a checklist you run yourself -- not a subagent dispatch.

1. Spec coverage: Skim each section/requirement in the design doc. Can you point to a task that implements it? List any gaps.

2. Placeholder scan: Search the plan for red flags -- any of the patterns from the "No Placeholders" section above. Fix them.

3. Type consistency: Do the types, method signatures, and property names used in later tasks match what was defined in earlier tasks? A function called clearLayers() in Task 3 but clearFullLayers() in Task 7 is a bug.

4. Dependency ordering: Can each task be executed after its dependencies complete? Are there circular dependencies? Is the ordering optimal for parallelization?

5. Command accuracy: Are the test commands, build commands, and file paths correct for this project's toolchain?

If you find issues, fix them inline. No need to re-review -- just fix and move on. If you find a spec requirement with no task, add the task.

---

Plan Review Swarm

After self-review, invoke the review swarm to scrutinize the implementation plan before the user approves it.

Automatic dispatch: Run the swarm for all medium-to-large features and improvements. Skip only for clearly trivial work.

INVOKE jig:review using the Skill tool with mode: plan. Pass the plan document path. The review skill discovers PLAN specialists (stage: plan or both), dispatches them in parallel with the full plan + PRD + section hints + codebase access. After the specialist swarm, a plan logic reviewer (Opus) performs deep correctness analysis. Findings are scored and returned as a unified report.

Present the swarm findings to the user before asking for approval:

"Plan written and self-reviewed. The review swarm found these concerns:"

{swarm report — including plan logic reviewer findings}

"Want to address any of these before approving the plan?"

If the user requests changes based on findings, update the plan and re-run the self-review checklist. Do not re-run the swarm unless the changes are substantial.

---

Plan Output

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

---

Execution Handoff

After saving the plan, offer the execution choice:

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

1. Team-Driven (parallel) -- Spawns implementer teammates in split panes, staggered review pipeline. Best for 3+ independent tasks touching different files.

2. Subagent-Driven (sequential) -- Fresh subagent per task, two-stage review after each. Best for coupled tasks or fewer than 3 tasks.

Which approach?"

If Team-Driven chosen:

• REQUIRED: Use team-dev
• Parallel implementation with spec compliance + code quality review gates

If Subagent-Driven chosen:

• REQUIRED: Use sdd
• Serial execution with two-stage review per task

Read jig.config.md for parallel-threshold and default-strategy to inform the recommendation, but always let the user choose.

---

Key Principles

• DRY -- do not repeat yourself across tasks (except code blocks, which must be self-contained)
• YAGNI -- do not add features not in the approved design
• TDD -- tests first, implementation second
• Frequent commits -- one commit per task minimum
• Zero ambiguity -- if an engineer has to guess, the plan failed

---

Integration

Called by:

• brainstorm (terminal state) -- after design is approved
• kickoff during the PLAN stage

Terminal state:

• Invoke team-dev (parallel) or sdd (sequential)

Related skills:

• brainstorm -- produces the design this skill consumes
• prd -- produces PRD with acceptance checklist referenced in plan header
• tdd -- implementers use TDD during execution
• team-dev -- parallel execution engine
• sdd -- sequential execution engine

---

Common Mistakes

Mistake	Consequence	Fix
Vague steps without code	Engineer guesses, builds wrong thing	Every code step has a complete code block
Missing file paths	Engineer creates files in wrong locations	Exact paths always, verify against project structure
Placeholders in test code	Tests do not actually test anything	Write real assertions with real expected values
Tasks too large	Context overload, errors compound	Each step is 2-5 minutes of focused work
Missing dependencies	Task fails because prerequisite not built	Declare blockedBy for every dependent task
Inconsistent naming across tasks	Runtime errors, undefined references	Self-review checks type consistency
Skipping self-review	Spec gaps ship, plans have contradictions	Always run the 5-point self-review
No PRD/design reference in header	Spec reviewers cannot find acceptance criteria	Include reference lines when documents exist