Project kickoff
A Claude skill that gives your projects persistent memory — interview-based kickoff with auto-tracked decisions and progress.
npx -y skills add magi12345/project-kickoff --skill project-kickoffAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Interview-style skill for starting a new project. It clarifies the goal and scope, then creates three tracking files (PROJECT.md, DECISIONS.md, PROGRESS.md) that act as persistent memory across all future Claude sessions. Use this skill EVERY time the user starts a new project of any kind — an app, a website, a plugin, a design, a prototype, a documentation or knowledge system, anything. Typical trigger phrases: I'm starting a new project, let's build something new, I want to make a website/app/plugin/tool, let's kick off, new project. At the end of the interview the skill doesn't just ask questions — it creates the three tracking files, writes the tracking convention into the project's CLAUDE.md, and from then on every conversation automatically records important decisions and completed units of work.
SKILL.md
9.4 KB, ~2.1k tokens by cl100k_base, as published. Nobody here has run it
Project Kickoff
Guides the user through clarifying goals, scope, and key decisions at the start of a new project, then creates three tracking files that serve as the project's memory for its entire lifetime — across every future Claude session.
Language: Conduct the entire interview and write all generated files in the language the user is speaking. If the user writes in Spanish, ask in Spanish and write Spanish PROJECT.md/DECISIONS.md/PROGRESS.md files. The headings shown in the templates below should be translated accordingly.
When to use
When the user is starting a new project — any kind: an app, a website, a plugin, a design or prototype, a tool, a knowledge base or documentation system.
Do NOT use it for a new component or feature inside an ongoing project — this is for kickoff only.
The three files
The output of this skill is three markdown files that live alongside the project:
1. PROJECT.md — What we're building
A snapshot of the project that records the initial state, but gets updated when scope or direction changes substantially. Contents:
# [Project name]
## What
One or two sentences describing what we're building.
## For whom / why
Target user or use case + motivation.
## V1 scope
- Concrete things that are definitely in
- As a list
## Out of scope
- Concrete things that are not
- As a list (marked as V2 if planned for later)
## Look & feel
One or two sentences about the visual / tonal direction (skip if not applicable).
## Stack / tools
What we're using to build it.
## Where it lives
The path of the three tracking files (folder).
2. DECISIONS.md — What changed and why
A chronological log, newest on top. Each entry: date + title + 1–3 sentence rationale. Auto-recorded — whenever an explicit decision comes up in conversation (scope change, change of technical direction, aesthetic change, discarded idea), Claude writes it in.
# Decision log
## 2026-06-09 — [Short descriptive title]
Brief rationale.
## 2026-06-08 — [Short descriptive title]
Brief rationale.
3. PROGRESS.md — What's been built
A chronological log, newest on top. Entries are added when a substantial unit is completed (a component, a feature, a screen) — not small tweaks and bug fixes. Auto-recorded — whenever a substantial unit is finished, Claude writes it in.
# Completed work
## 2026-06-09 — [Unit name]
- What it does
- What variants it has
- Anything a future session needs to know
## 2026-06-08 — [Unit name]
...
Process
1. Opening question
Start with a single question, no clarifying questions yet:
What would you like to build? Give me 1–2 sentences.
2. Clarifying questions
Only ask about what the opening answer did not already cover. Don't ask questions whose answers you already have. Adapt the wording and depth of these questions to the kind of project the user described.
- Who is it for / what's the use case
- What's in V1, what's left out (or pushed to V2)
- Look & feel direction in one sentence (skip if irrelevant for this project)
- Stack / tools, if applicable (and whether there's an existing system or library to build on, or it's from scratch)
- Where the three tracking files will live (project folder path)
3. PROJECT.md proposal
Based on the answers, write a PROJECT.md proposal following the template above and ask for approval. Do not create any files until the user approves.
If the user requests changes, revise and show it again.
4. Creating the three files
After approval:
-
Create
PROJECT.mdat the agreed path. -
Create
DECISIONS.mdempty except for the header and a first entry:# Decision log ## [date] — Project started Initial scope and direction recorded in PROJECT.md. -
Create
PROGRESS.mdempty except for the header:# Completed work _No entries yet._ -
Write the convention into the project's
CLAUDE.md(create it if it doesn't exist). This step is critical: without it, a future fresh session won't know about the tracking files and auto-recording will silently stop. The block to add:## Project tracking files (project-kickoff convention) Three tracking files live alongside this project: `PROJECT.md`, `DECISIONS.md`, `PROGRESS.md`. Maintain them automatically in every conversation: - `DECISIONS.md` — on every explicit decision, add a new entry ON TOP (date + short title + 1–3 sentence rationale). A decision = scope change, change of direction, aesthetic change, discarded idea. No minor technical details. If a decision overturns a previously recorded direction or assumption, use the richer narrative format (What we thought / Why we turned / What changed). - `PROGRESS.md` — when a substantial unit (component, feature, screen) is completed, add a new entry ON TOP. No small tweaks or bug fixes. - `PROJECT.md` — update it when scope or direction changes substantially (and record that change in DECISIONS.md as well). After recording, mention it in one sentence ("Recorded in DECISIONS.md: ..."). Only ask when uncertain (thinking out loud ≠ a decision).
5. Wrap-up
Tell the user:
- The three files have been created, and where to find them.
- The convention is now in
CLAUDE.md, so every future conversation will automatically write toDECISIONS.mdon important decisions and toPROGRESS.mdwhen larger units are completed. - If the scope or direction changes substantially,
PROJECT.mdgets updated too (and the change also shows up inDECISIONS.md).
Auto-recording rules
After this skill completes, in all further conversations on the project:
When to write to DECISIONS.md
Whenever an explicit decision is made in conversation that departs from the previous state or sets a meaningful new direction:
- Scope change (something dropped from V1, something pulled forward from V2, a new requirement)
- Change of technical direction (e.g. "let's drop Tailwind, vanilla CSS instead")
- Aesthetic change (e.g. "dark background after all")
- Naming / branding change
- A discarded idea worth writing down so it doesn't have to be re-debated later
- Anything else that would help a future session answer "why is it like this?"
Do not record minor technical details (class names, padding values) — those live in the code.
Entry format: date + short title + 1–3 sentence rationale. Newest on top.
Direction reversals get the richer format. When a decision overturns a previously recorded direction or assumption — not just adjusts it, but reverses it — use the narrative entry format, because the reasoning behind a turn is worth more than a bare statement:
## [date] — [Title of the turn]
**What we thought**: ...
**Why we turned**: ...
**What changed**: ...
For ordinary decisions, stick to the short format — don't pad small decisions into the narrative structure.
When a decision also affects the contents of PROJECT.md (e.g. something drops out of V1), update PROJECT.md too, don't just log it in DECISIONS.md.
When to write to PROGRESS.md
When a substantial unit is completed:
- A component
- A feature (e.g. a command, a page)
- A screen (for design projects)
- Anything a future session will need to know "this already exists"
Do not record small tweaks, bug fixes, or styling refinements.
Entry format: date + unit name + 2–5 lines about what it does, what variants it has, and anything a future Claude instance should know about it.
How to record
Don't ask every time ("Should I add this to DECISIONS?"). If the decision is clear and the user has confirmed it, just write it in and mention it in one sentence:
"Recorded in
DECISIONS.md: [date] — [title]."
Or:
"Added to
PROGRESS.md: [unit name]."
Only ask when uncertain ("Should I record this?"). The typical uncertain case: the user is thinking out loud and it's not clear whether the option they voiced is a decision yet.
Process overview
Trigger (starting a new project)
↓
Opening question: "What would you like to build?"
↓
Ask only about what isn't known yet
(adapt questions to the kind of project)
↓
PROJECT.md proposal → approval
↓
Create the three files in the agreed folder:
- PROJECT.md (filled in)
- DECISIONS.md (empty + opening entry)
- PROGRESS.md (empty)
+ convention block in the project's CLAUDE.md
↓
Wrap-up: explain auto-recording
↓
[in all further conversations on the project, Claude
keeps the three files up to date based on the
CLAUDE.md convention — even in fresh sessions]
Gives 0 of the 12 instructions most memory context skills give in ~2.1k tokens
Counted across 674 of the 847 authors here whose files we hold, read 2026-08-06
- inform the user when setup is completein 21 of 674, across 6 files
- confirm the draft with the user before writingin 21 of 674, across 6 files
- update the agent skills block in place if it existsin 21 of 674, across 6 files
- present findings to the userin 20 of 674, across 5 files
- write the three docs files from seed templatesin 20 of 674, across 5 files
- ask the user about each decision one at a timein 19 of 674, across 4 files
- edit CLAUDE.md if it existsin 18 of 674, across 3 files
- explore current repo statein 18 of 674, across 3 files
- do not overwrite user edits to surrounding sectionsin 18 of 674, across 3 files
- back up the original file before overwritingin 16 of 674, across 8 files
- keep the memory index under 200 linesin 15 of 674
- Provide actionable steps and verificationin 13 of 674, across 2 files
Said here and by no other author read
- ask one opening question
- conduct the interview in the user's language
- write all generated files in the user's language
- propose the project file for approval
- create the three tracking files after approval
- write the tracking convention into the memory file
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.