Writing Plans
Write implementation plans that let a skilled engineer work safely with almost
no prior context. A plan names the exact files, interfaces, behavior, tests,
documentation, and commands needed for each independently verifiable task.
Plans are durable repository artifacts, not chat-only checklists. Each task
also ends with a compile, diff review, and conventional commit gate.
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: If working in an isolated worktree, use the repository's worktree
workflow at execution time. Do not make the plan depend on chat history.
Save plans to: .agents/plans/YYYY-MM-DD-<feature-name>.md
Every plan is Markdown with YAML frontmatter. The frontmatter is the
machine-readable index for cross-session tracking:
yaml
---
id: YYYY-MM-DD-<feature-name>
name: <Feature Name> implementation plan
kind: plan
category: implementation
status: pending
description: <one-sentence purpose>
---
Use these statuses:
| Status | Meaning | Transition |
|---|
| Plan is being written or self-reviewed | → , |
| Plan is complete and ready to implement | → , |
| Implementation has started | → , , |
| A documented decision or missing input prevents progress | → , , |
| Implementation and verification are complete | terminal |
| Replaced by a newer plan; link the replacement | terminal |
Start as
, change to
only after self-review, and change to
when implementation begins. Change to
only after verification.
Use
for a concrete missing decision or input and
when a
newer plan replaces it. Update the same plan file; never create a second status
file.
To track plans:
bash
find .agents/plans -maxdepth 1 -type f -name '*.md' -print | sort
rg -n '^status:' .agents/plans
rg -l '^status: (pending|running|blocked)$' .agents/plans
Use
when work must survive a worker, session, or context
boundary. Use
code-review-axes-and-quality
after implementation when a
complete quality review is needed. Keep these as separate stages: planning
defines the contract, implementation proves it, and review checks the result.
Scope Check
If the spec covers multiple independent subsystems, split it into separate
plans. Each plan must produce a working, testable increment on its own. Do not
hide unrelated cleanup inside an implementation plan.
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.
- Design units with clear boundaries and well-defined interfaces. Each file should have one clear responsibility.
- You reason best about code you can hold in context at once, and your edits are more reliable when files are focused. 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, don't unilaterally restructure - but if a file you're modifying has grown unwieldy, including a split in the plan is reasonable.
This structure informs the task decomposition. Each task should produce self-contained changes that make sense independently.
Task Right-Sizing
A task is the smallest unit that carries its own test cycle and is worth a
fresh reviewer's gate. When drawing task boundaries: fold setup,
configuration, scaffolding, and documentation steps into the task whose
deliverable needs them; split only where a reviewer could meaningfully
reject one task while approving its neighbor. Each task ends with an
independently testable deliverable.
Bite-Sized Task Granularity
Each step is one action, normally 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
- "Compile the affected project or package" - step
- "Inspect the complete diff and run " - step
- "Commit the verified increment with a conventional message" - step
Plan Document Header
Every plan MUST start with this frontmatter and header:
markdown
---
id: YYYY-MM-DD-<feature-name>
name: [Feature Name] implementation plan
kind: plan
category: implementation
status: draft
description: [One-sentence purpose]
---
# [Feature Name] Implementation Plan
> **For agentic workers:** Use `delegation-mode` for resumable or multi-worker execution. Steps use checkbox (`- [ ]`) syntax for tracking.
**Goal:** [One sentence describing what this builds]
**Architecture:** [2-3 sentences about approach]
**Tech Stack:** [Key technologies/libraries]
**Spec:** [path to the spec/design doc this plan implements — the plan
argues from the spec, so the spec travels with it; executors read both]
## Global Constraints
[The spec's project-wide requirements — version floors, dependency limits,
naming and copy rules, platform requirements — one line each, with exact
values copied verbatim from the spec. Every task's requirements implicitly
include this section.]
---
Task Structure
markdown
### 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`
**Interfaces:**
- Consumes: [what this task uses from earlier tasks — exact paths and signatures]
- Produces: [what later tasks rely on — exact paths, names, parameters, and
return types. This block is the contract for an implementer who sees only
this task.]
- [ ] **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: Compile the affected project or package**
Run the narrowest real compile/build command that proves the changed
interfaces are valid. Record the exact command and result in the task handoff.
```bash
<project compile or build command>
```
Expected: exit 0 with no compile errors.
- [ ] **Step 6: Review the complete diff**
```bash
git diff --check
git diff -- <files owned by this task>
git status --short
```
Confirm there are no unrelated files, generated artifacts, secrets, or
unreviewed edits. If the diff is not scoped, fix it before committing.
- [ ] **Step 7: Commit the verified increment**
```bash
git add tests/path/test.py src/path/file.py
git diff --cached --check
git diff --cached
git commit -m "feat: add specific feature"
```
Record the commit hash and verification evidence in the task result. Do not
mark the task `done` before the compile, diff, and commit gates pass.
No Placeholders
Every step must contain the actual content an engineer needs. These are plan failures — never write them:
- "TBD", "TODO", "implement later", "fill in details"
- "Add appropriate error handling" / "add validation" / "handle edge cases"
- "Write tests for the above" (without actual test code)
- "Similar to Task N" (repeat the code — the engineer may be reading tasks out of order)
- Steps that describe what to do without showing how (code blocks required for code steps)
- References to types, functions, or methods not defined in any task
Self-Review
After writing the complete plan, review it against the spec with fresh eyes.
This is a checklist you run yourself, not a subagent dispatch.
1. Spec coverage: Skim each section/requirement in the spec. Can you point to a task that implements it? List any gaps.
2. Placeholder scan: Search your 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 you used in later tasks match what you defined in earlier tasks? A function called
in Task 3 but
in Task 7 is a bug.
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.
Execution Handoff
After saving the plan, offer execution choice. Keep the path in the repository's
namespace:
"Plan complete and saved to .agents/plans/<filename>.md
1. Delegated execution (recommended) — Dispatch a fresh worker per task,
record each result under
, and review between tasks.
2. Inline execution — Execute tasks in this session with checkpoints and
update the plan status as work progresses.
Which approach?"
For delegated execution, use
: stable task IDs, one result
file per task, explicit resume state, and a final aggregation only after every
task is confirmed. For inline execution, keep the same task boundaries and
update checkboxes and status in the plan.
Before marking the plan or implementation complete, require fresh evidence:
run the relevant compile/build and tests, inspect the full diff, run
, stage only the intended files, inspect the staged diff, and
commit the verified increment. Do not mark
from intent or an agent's
success report alone. A task without a recorded compile result, diff review,
and commit hash is incomplete.