initiating-a-project
Skill1 starsChanged 4 months agoMITView on GitHub
Turn a human-written product brief into a complete PRD and a populated harness, so a brand-new project goes from "rough idea" to "ready to write specs." Use at the very start of a new project (empty scaffold, placeholders still in AGENTS.md), when the user hands you a product brief, or when they say "start a new project," "here's my idea," or "turn this into a PRD." Ask the human for missing critical information rather than inventing it.
In this repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- .cursor/rules/00-harness.mdcCursor rule
- brainstormingSkill
- checkpointing-progressSkill
- executing-plansSkill
- idea-reviewSkill
- onboardingSkill
- <kebab-case-folder-name>Skill
- test-driven-developmentSkill
- verifySkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 30
Ask the human: when a missing answer changes *what gets built*: the differentiator, target users, the MVP scope boundary, hard constraints …
- Distinctive: line 31
Assume and state it: when there's a conventional default and the choice is low-stakes or easily reversed (a doc format, a naming …
- Very specific: line 43
Keep the PRD a living document — when product direction shifts later, edit it in place and append to its ## Changelog (same lifecycle as …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skill: Initiate a project (product brief → PRD → populated harness)
This is the front door for a brand-new project. It bridges a human's rough product brief to the point where the normal harness loop (spec → plan → execute) takes over. It's inspired by the "brief → PRD → build" method, implemented natively — no external framework, no dependency.
The flow
- Get the brief. The human copies
docs/product/_TEMPLATE-product-brief.mdtodocs/product/product-brief.mdand fills in what they know. If they pasted an idea into chat instead, capture it into that file first — the brief must exist on disk, because it's the durable input a later provider can re-read. - Read the brief and find the gaps. Compare it against what a PRD requires (checklist below). List every gap.
- Ask the human about critical gaps — don't invent them. See "Ask vs. assume".
- Write the PRD to
docs/product/PRD.mdfrom_TEMPLATE-prd.md. - Populate the harness from the PRD (see "Populate the harness").
- Hand off to the normal loop: pick the first feature/slice from the PRD, write its spec (
docs/specs/) and plan (docs/plans/), and execute. Everything downstream already exists in this harness.
What a PRD must contain (the required-information checklist)
A PRD is "done" only when each of these is answered — this list is exactly what makes a spec writable:
- Problem & why now — the pain, who has it, why it's worth solving now.
- Target users / personas — who this is for, concretely.
- The one differentiator — the single thing this product must do better than the alternatives; its reason to exist. (This usually becomes invariant #1 in
AGENTS.md.) - Goals & success metrics — how you'll know it worked (measurable where possible).
- Scope: in / out — the MVP boundary and explicit non-goals.
- Key features / capabilities — the handful that deliver the differentiator, roughly prioritized.
- Constraints — platform, tech stack, timeline, budget, compliance, team.
- Assumptions & risks — what you're betting on and what could sink it.
- Open questions — anything still unresolved (carry these forward, don't hide them).
Ask vs. assume
- Ask the human when a missing answer changes what gets built: the differentiator, target users, the MVP scope boundary, hard constraints (platform/compliance/deadline), success metrics. Batch these into one clear list rather than dribbling them out one at a time.
- Assume and state it when there's a conventional default and the choice is low-stakes or easily reversed (a doc format, a naming convention). Record the assumption in the PRD's "Assumptions" section so it stays visible and correctable.
- Never fabricate a metric, a user, or a differentiator to fill a blank. A confident-but-wrong PRD produces confident-but-wrong specs. A visible "OPEN QUESTION" beats an invented answer every time.
Populate the harness (from the PRD)
Once the PRD exists, use it to make the scaffold real:
AGENTS.md— fill in<PROJECT_NAME>, "What this project is" (from problem + differentiator), "Non-negotiable invariants" (the differentiator is usually invariant #1), "Where the code lives", "Quickstart" (build/test/run commands), and "Conventions in one breath".docs/system/— seed one capability file per major intended capability. They describe intended behavior now and become current behavior as features ship.docs/DECISIONS.md— log any foundational choices already made (stack, platform, architecture direction) as ADR-002 onward.docs/state/CURRENT.md— set the first active task (usually "write the spec for<first feature>").
Conventions to keep
- The brief is the human's; the PRD is yours to draft but theirs to approve. Surface open questions and get a nod before treating the PRD as settled.
- Keep the PRD a living document — when product direction shifts later, edit it in place and append to its
## Changelog(same lifecycle as specs; seecheckpointing-progress). - Brainstorming and idea-review are optional downstream — reach for
brainstormingwhen a feature's shape is unclear andidea-reviewwhen an idea is risky, but neither is a required gate between the PRD and a spec.
Output
A complete docs/product/PRD.md (approved by the human), a populated AGENTS.md, a seeded docs/system/, and docs/state/CURRENT.md pointing at the first spec to write.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

