write-todos
Write clear, actionable todos that workers can execute without losing architectural intent. Use when "create todos", "write todos", "break into tasks", "plan todos", "make todos", or creating work items from a plan. Ensures each todo has unambiguous expected outcomes, concrete examples, and explicit constraints so workers don't drift from the design.
Best use case
write-todos is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Write clear, actionable todos that workers can execute without losing architectural intent. Use when "create todos", "write todos", "break into tasks", "plan todos", "make todos", or creating work items from a plan. Ensures each todo has unambiguous expected outcomes, concrete examples, and explicit constraints so workers don't drift from the design.
Teams using write-todos should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/write-todos/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How write-todos Compares
| Feature / Agent | write-todos | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Write clear, actionable todos that workers can execute without losing architectural intent. Use when "create todos", "write todos", "break into tasks", "plan todos", "make todos", or creating work items from a plan. Ensures each todo has unambiguous expected outcomes, concrete examples, and explicit constraints so workers don't drift from the design.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Write Todos
Write todos that a worker agent can execute without access to the planning conversation. Every todo must be **self-contained** — a worker reading only the todo and the plan artifact must produce the correct result.
## Why This Matters
Workers implement exactly what's described. If a todo contains a code sketch using plain classes, the worker builds plain classes — even if the plan says "use functional code for everything." Architectural intent that only lives in the plan's prose gets lost. Every constraint must be **in the todo body itself**.
## Todo Structure
Every todo body follows this structure:
```markdown
**Plan:** `plans/YYYY-MM-DD-<name>.md`
## What
[One paragraph: what this todo produces and why it matters]
## Constraints
- [Explicit architectural constraints that MUST be followed]
- [Libraries/patterns to use — or explicitly NOT use]
- [Reference existing code patterns: "Follow the pattern in src/foo.ts"]
## Files
- `src/path/to/file.ts` — [what this file does]
- `src/path/to/other.ts` — [what this file does]
## Expected Outcome
[Concrete description of what the finished code looks like]
### Example
[Short code snippet showing the expected shape — imports, key patterns, structure]
## Acceptance Criteria
- [ ] [Specific, verifiable criterion]
- [ ] [Another criterion]
- [ ] [Build/lint/test passes]
```
## Rules
### 1. Constraints Are Explicit, Not Implied
If the plan says "use Effect v4 for all services," every service todo must repeat that:
| Bad (implicit) | Good (explicit) |
|---|---|
| "Build the EventBus service" | "Build the EventBus as an Effect v4 service. Import from `effect`. Use `Effect.gen`, `Layer`, and `Context.Tag` — not plain classes." |
| "Add WebSocket support" | "Add WebSocket support using the `ws` package. Do NOT use `socket.io`." |
| "Create the component" | "Create the component using React 19 + Tailwind v4 utility classes. No CSS modules, no styled-components." |
### 2. Examples Show The Real Shape
Include a short code snippet showing the expected import style, patterns, and structure. This is the single most effective way to prevent drift.
```markdown
### Example
The service should look like this (not a plain class):
\```typescript
import { Effect, Context, Layer } from "effect"
class EventBus extends Context.Tag("EventBus")<EventBus, {
readonly subscribe: (topic: string) => Effect.Effect<Subscription>
readonly publish: (event: PiEvent) => Effect.Effect<void>
}>() {}
const EventBusLive = Layer.effect(EventBus, Effect.gen(function* () {
// ... implementation using Effect primitives
}))
\```
```
Without examples, workers default to the most common pattern they know — which is usually plain TypeScript classes.
### 3. Anti-Patterns Are Named
If there's a wrong way that looks right, call it out:
```markdown
## Constraints
- Use Effect v4 services with `Context.Tag` and `Layer`
- **Do NOT** use plain classes with manual observer patterns (no `new EventBus()`, no `Set<() => void>` listener tracking)
- **Do NOT** use `useSyncExternalStore` with hand-rolled subscribe — use Effect's reactive primitives
```
### 4. Each Todo Is Self-Contained
A worker reads: (1) the todo body, (2) the plan artifact, (3) existing code. That's it. They don't read other todos. So:
- Reference the plan path in every todo
- List all files to create or modify
- Note which existing files to read for context
- Include any conventions discovered during planning
### 5. Todos Are Sequenced
Number todos and note dependencies:
```markdown
**Title:** "Todo 3: Build EventNode state machine"
**Body includes:** "Depends on Todo 2 (types in `src/core/types.ts`). Read that file first."
```
### 6. Size Is Right
Each todo should be **one focused unit of work** — a worker can complete it in one session and make one commit. If a todo has more than 3 files to create, consider splitting it.
### 7. Acceptance Criteria Are Verifiable
Every criterion should be checkable by running a command or reading the output:
| Bad (vague) | Good (verifiable) |
|---|---|
| "Code is clean" | "`vp check` passes with no errors" |
| "Works correctly" | "Running `node -e 'import { EventBus } from \"./src/services/EventBus\"'` succeeds" |
| "Tests pass" | "`vp test src/core/EventNode.test.ts` passes" |
| "Follows conventions" | "All imports use `effect` package, no plain class instantiation" |
## Checklist Before Creating Todos
Before calling `todo(action: "create")`, verify:
- [ ] Every architectural decision from the plan appears as an explicit constraint in at least one todo
- [ ] Every todo has a code example showing expected shape (imports, patterns, structure)
- [ ] No todo relies on context only available in the planning conversation
- [ ] Anti-patterns are named in relevant todos ("do NOT use X")
- [ ] Todos are numbered and dependencies noted
- [ ] Acceptance criteria are verifiable commands, not subjective judgmentsRelated Skills
skill-creator
Create new agent skills following the Agent Skills specification. Use when asked to "create a skill", "add a new skill", "write a skill", "make a skill", "build a skill", or scaffold a new skill with SKILL.md. Guides through requirements, planning, writing, registration, and verification.
session-reader
Efficiently read and analyze pi agent session JSONL files. Use when asked to "read a session", "review a session", "analyze a session", "what happened in this session", "load session", "parse session", "session history", "go through sessions", or given a .jsonl session file path.
self-improve
End-of-session retrospective that identifies improvements to agent config, tests, docs, and code. Use when asked to "self-improve", "reflect on session", "what can we improve", "session retrospective", "end of session review". Creates actionable todos from findings.
presentation-creator
Create data-driven presentation slides using React, Vite, and Recharts with Sentry branding. Use when asked to "create a presentation", "build slides", "make a deck", "create a data presentation", "build a Sentry presentation". Scaffolds a complete slide-based app with charts, animations, and single-file HTML output.
learn-codebase
Discover project conventions and surface security concerns. Use when starting work in a new or unfamiliar project, when asked to "learn the codebase", "check project rules", "what are the conventions", "onboard to this project", or "anything shady in this codebase". Scans agent config files (.claude/, .cursor/, CLAUDE.md, etc.) and runs a security/smell sweep for hardcoded secrets, insecure patterns, suspicious dependencies, and dangerous configurations.
iterate-pr
Iterate on a PR until CI passes. Use when you need to fix CI failures, address review feedback, or continuously push fixes until all checks are green. Automates the feedback-fix-push-wait cycle.
github
Interact with GitHub using the `gh` CLI. Use `gh issue`, `gh pr`, `gh run`, and `gh api` for issues, PRs, CI runs, and advanced queries.
frontend-design
Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.
commit
Read this skill before making git commits
code-simplifier
Simplifies and refines code for clarity, consistency, and maintainability while preserving all functionality. Use when asked to "simplify code", "clean up code", "refactor for clarity", "improve readability", or review recently modified code for elegance. Focuses on project-specific best practices.
cmux
Manage terminal sessions and browser surfaces via cmux — spawn workspaces for dev servers, test runners, background tasks, and embedded browsers. Read output, send commands, interact with web pages, and orchestrate multi-terminal workflows.
add-mcp-server
Add an MCP server to pi. Use when asked to "add mcp server", "configure mcp", "add mcp", "new mcp server", "setup mcp", "connect mcp server", or "register mcp server". Handles both global and project-local configurations.