project
Structured project workflow: brainstorm, spec, plan, execute multi-step tasks
Best use case
project is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Structured project workflow: brainstorm, spec, plan, execute multi-step tasks
Teams using project 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/project/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How project Compares
| Feature / Agent | project | 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?
Structured project workflow: brainstorm, spec, plan, execute multi-step tasks
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
## Project Workflow Use projects for complex multi-step tasks. Each project follows a lifecycle: **brainstorm → spec → plan → execute → done**. ### How to use 1. Call `project_create` to start a new project 2. Call `project_next_task` — it tells you what to do 3. Do what it says (the turn will end automatically after key actions) 4. When the user responds, continue from where you left off 5. Repeat until done **Two key tools drive the workflow:** - `project_next_task` — tells you what to do NOW (does not advance phases) - `project_task_done` — signals you're done and advances to the next phase ### Spec vs Plan - **Spec** = WHAT and WHY. This is the problem definition. Describes the goal, requirements, constraints, and acceptance criteria. Does not list steps or say how to do it. - **Plan** = HOW. This is the implementation plan to solve the problem described in the spec. A numbered checklist of concrete action steps to implement the spec. Written AFTER the spec is approved. These are separate phases. Do not include a plan in the spec or a spec in the plan. ### Key rules - **Do the work, then call project_task_done** — don't skip ahead - **During brainstorming:** Ask questions to understand the project, then write the spec (WHAT, not HOW) with `project_update_spec` - **Review phases:** Read the user's response. If they approve, call `project_task_done`. If they have feedback, revise the artifact. - **During planning:** Write a concrete step-by-step plan (HOW) with `project_update_plan` - **During execution:** Work through steps freely — mark in_progress, do work, mark done, repeat - **Write output files to the project directory** ### Command handling When invoked as `!project` or `/project`: - No args or `status`: call `project_list` or `project_status` - `create <description>`: call `project_create` - `status <slug>`: switch to that project and call `project_status` - `list`: call `project_list` User said: $ARGUMENTS
Related Skills
vault
Unified knowledge base — shared Obsidian-compatible vault for curated pages, journal entries, and user notes
tabstack
Your primary tool for any web, PDF, or research task. More powerful than web_fetch — prefer this for all research, web reading, and data extraction. Triggers on: 'tell me about,' 'what is,' 'look up,' 'find out,' 'research,' 'summarize this article,' 'read this PDF,' 'check this site,' 'what does this page say,' 'extract data from,' 'find the price on,' 'compare X vs Y,' 'is it true that,' or any URL/link. Handles JavaScript-heavy websites, PDFs, structured data extraction, content transformation, multi-source research with citations, and multi-step browser automation.
postmortem
Structured blameless analysis of what went wrong in this conversation — identifies root causes and proposes minimal, specific fixes
newsletter
Compose and deliver a narrative newsletter summarizing autonomous agent activity in the window.
mcp
Admin tools for inspecting and restarting connected MCP servers — status, resources, prompts. Does NOT expose tools provided by MCP servers; those appear as mcp__server__tool and are fetched via tool_search.
ingest
Fetch a URL, workspace file, or attachment and integrate its content into the vault — one primary page plus cross-linked updates to related pages
health
Show agent diagnostic status — process, MCP, heartbeat, tools, embeddings
garden
Vault gardening sweep — merge, link, split, and tidy agent pages
dream
Review recent journal entries and conversations, distill insights into vault pages
claude_code
Delegate coding tasks to Claude Code as a subagent. Use when asked to fix bugs, add features, refactor code, write tests, review code, or do any work that requires reading and editing files in a codebase. Triggers on: 'fix this bug', 'add a feature', 'refactor', 'write a test', 'review this code', 'update the config', 'clean up', or any request involving code changes in a repository.
background
Start, monitor, and stop long-running background processes (servers, watchers, builds). Returns immediately with a job ID you can poll for output.
writing-clearly
Edit prose for clarity and concision using Strunk's *The Elements of Style* (1918). Use whenever you have a draft — documentation, commit messages, blog posts, replies — that should be tightened before it goes out.