agentleFS
ridwankustantoridwankustanto/mandor/.claude/skills/pm-new/SKILL.md

pm-new

Skill2 starsChanged 41 days agoMITView on GitHub

Phase 0, Intent. Start a project by interviewing the user with form-style questions until the objective, the success signal, the non-goals and the point of view are unambiguous. Creates project.db, the product overview and standards/point-of-view.md. Use when starting a new project, when the user says "new project", "start a project", "/pm-new", or when project.db does not exist yet.

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 6 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Ask everything with AskUserQuestion: Never a wall of prose. Four questions per form at most, recommendation first and labelled.

    line 11
  2. Distinctive:

    Read the round 1 answers. Ask only about what is still ambiguous.

    line 36
  3. Very specific:

    If we do not have one, the AI gives us one, and its one is the most probable answer.

    line 47
  4. Distinctive:

    Ask with forms, drafted options first, the user corrects:

    line 51
  5. Distinctive:

    Write back, in your own words and under 150 words:

    line 68
  6. Very specific:

    Do not build the work tree yet.

    line 92
  7. Very specific:

    Write the confirmed non-goals into the overview verbatim.

    line 95
  8. Very specific:

    Only after the user confirmed the restatement.

    line 105
  9. Distinctive:

    Do not propose a solution, a stack, or a screen in this phase.

    line 111
  10. Very specific:

    Do not write epics or stories.

    line 112
  11. Distinctive:

    Do not accept "make it good" or "like <product> but better" as a success signal.

    line 113
  12. Distinctive:

    Ask again, with options.

    line 114

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/pm-new: get the objective crystal clear

The whole point of this phase is that nobody downstream has to guess. You are not filling a form, you are interrogating an idea until it holds still.

Ask everything with AskUserQuestion. Never a wall of prose. Four questions per form at most, recommendation first and labelled.

Round 1, the fixed seven

Two forms, four then three. Do not ask them as prose.

  1. What are we building, in one sentence?
  2. Who is it for, and what do they do today instead?
  3. What breaks, or stays broken, if this never exists?
  4. What does done look like for the first usable version?
  5. Category: whichever buckets the owner sorts projects into (personal, work, client, game). Offer the ones already used in other projects if you can see them.
  6. Stack: known already, or open for /pm-discover to decide?
  7. Full or lite? Recommend lite for a one or two session build, a prototype that may be thrown away, or an internal tool for one user. Recommend full for anything with real users, more than one milestone, or a launch. Lite skips discover, spec, flows and design. It keeps the point of view and the edit pass, because a fast build is exactly where a burrito gets called lunch.

For 1 and 4, offer drafted options rather than a blank box. People correct a wrong draft faster than they write a right one from nothing.

Round 2, close the gaps

Read the round 1 answers. Ask only about what is still ambiguous. Typical gaps:

  • A scope boundary that could reasonably be read two ways
  • A success signal with no number or no observable event
  • A user group that is actually two groups with different needs
  • A constraint the user mentioned in passing and has not made explicit

Generate these questions from the answers, not from a checklist.

Round 3, the point of view

If we do not have one, the AI gives us one, and its one is the most probable answer. This round fills standards/point-of-view.md, which every agent reads before building. It runs in lite mode too, and it is the one round that is never skipped.

Ask with forms, drafted options first, the user corrects:

  1. Who do we want to be to the user? Not what it does, what it is to them.
  2. What do they care about that they would not think to say? Offer what you noticed.
  3. Three beliefs that shape decisions, specific enough that someone could disagree.
  4. Three things we refuse to be. Name the generic version of this product precisely.
  5. How should it feel, in three adjectives that exclude something.
  6. References they admire, inside this category and outside it, and what to take from each. Push for one from outside software.
  7. The detail nobody asked for: one example of the care we want users to find.

Reject answers any product could give. "Simple and intuitive" is a no, ask again with sharper options. The test: would a competitor write the same sentence? Then it is not a point of view.

Round 4 and onward, restate and confirm

Write back, in your own words and under 150 words:

  • The objective
  • The success signal, observable and specific
  • The top three non-goals
  • Who we are to the user, and the one thing we refuse to be

Then ask, with the form UI, whether that restatement is right. Offer: "Yes, that is it", "Close, one thing is wrong", "No, start that part over".

Loop until the user picks "Yes, that is it". This is the exit condition. Not the form being filled. Not you feeling confident. The user confirming your restatement.

Then write it down

./pm init "<Project Name>"            # add --lite if round 1 chose lite
  • Fill docs/product/overview.md from the template: objective, success signal, the three non-goals, users, constraints. Helicopter view only, stays short.
  • Fill standards/point-of-view.md from round 3, then delete the pm:unfilled marker on its first line. Until it is gone, ./pm check fails past intent.
  • Fill the Context block at the top of CLAUDE.md.
  • Do not build the work tree yet. Epics and stories are /pm-spec, and creating them now means committing to a shape before the research that should inform it.

Write the confirmed non-goals into the overview verbatim. They become constraint stories with priority=wont in /pm-spec, which is how scope creep gets caught later.

Close the gate

./pm gate approve intent "objective restated and confirmed by <user>"
./pm phase next

Only after the user confirmed the restatement. If they want to move on while something is still fuzzy, that is their call, but record it: ./pm gate skip intent "<reason>".

Do not

  • Do not propose a solution, a stack, or a screen in this phase. Intent only.
  • Do not write epics or stories. They come in /pm-spec, after research.
  • Do not accept "make it good" or "like <product> but better" as a success signal. Ask again, with options.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.