plan-intake
Skill2 starsChanged 10 days agoMITView on GitHub
Interview the user in their language (and read what already exists) to write a product's context in docs/context — problem, users, domain, constraints, environments, and every open gap with owner and question. Use when the user says "entrevistame", "intake", "relevamiento", "contexto del proyecto", "tengo una idea", "armá el contexto", "gather requirements", when project-new reaches gate 1, or when a new feature needs context the project doesn't have. Only asks what the house harness doesn't already answer.
In this repository18 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .claude/CLAUDE.mdCLAUDE.md
- template/CLAUDE.mdCLAUDE.md
- build-storySkill
- check-bugbashSkill
- check-healthSkill
- check-securitySkill
- coordinateSkill
- harness-upgradeSkill
- keelokitSkill
- plan-backlogSkill
- project-adoptSkill
- project-dashboardSkill
- project-newSkill
- project-reportSkill
- run-localSkill
- ship-releaseSkill
- ship-setupSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 2 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 12
Read before starting: references/question-bank.md and references/context-format.md.
- Distinctive: line 16
Ask once what already exists in writing and where (briefs, notes, decks, screenshots, a repo, links).
- Very specific: line 18
Read each source against the question bank.
- Very specific: line 44
Write the five files exactly as references/context-format.md shows: product.md, domain.md, constraints.md, environments.md, gaps.md.
- Very specific: line 47
Never invent: No source, no fact: write [GAP-nnn] inline and add the row to gaps.md (file, what's missing, owner, exact question, blocking …
- Very specific: line 49
Every fact carries its source id.
- Very specific: line 56
No vague phrases ("robust", "user-friendly", "best practices"…) — python3 .keelokit/bin/doctor.py rejects them.
- Distinctive: line 57
No secrets, no personal data of real people.
- Very specific: line 61
If the project has .keelokit/profile.toml, update its traits from what the context now says: personal-data when it stores anything that …
- Very specific: line 69
If the project has .keelokit/bin/doctor.py, run it and fix every docs/context error.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Intake — only what this product needs, never invented
The house harness (stack, coding, testing, security, CI, agent protocol) is already decided and inherited. This interview covers only the product: why it exists, who uses it, its domain, its hard limits and its environments. Target: 15–25 questions, fewer when documents exist.
Read before starting: references/question-bank.md and references/context-format.md.
1. Harvest before asking
- Ask once what already exists in writing and where (briefs, notes, decks, screenshots, a repo, links). List what you found and confirm the list before reading.
- Read each source against the question bank. Pre-fill every question it answers, citing the
source id (
S1,S2…). A document that is undated, a draft, or contradicts another source is to confirm, not answered. - Existing repo: read the README, manifests, CI and top-level folders (structure and commands, not logic). What you derive is inferred until the user confirms it.
- Show the delta before any question:
Sources: 3 (S1–S3) · Answered by documents: 11/26 · To confirm: 4 · To ask: 11
2. Interview the delta
- At most 4 questions per turn, with the multiple-choice tool when the answer space is known, always with a free-text path and an "I don't know yet" option that records a gap.
- Accept only measurable answers: a number, a name, a date, a regulation, a concrete example. Re-probe a vague one once — "how would we know it holds?", "what number?", "what happens if it doesn't?" — then record a gap instead of writing the vague version.
- High-signal probes: the last time this problem hurt (a concrete story), what they use today and why it fails, what would make them abandon the product, the rule that must never break.
- Separate wish from requirement: "is it required to launch, or where you want to get to?" Wishes go to the PRD's later scope, not to constraints.
- Three unanswered in a row in one block → stop the block, record gaps with the right owner, move on.
3. Write docs/context
Write the five files exactly as references/context-format.md shows:
product.md, domain.md, constraints.md, environments.md, gaps.md.
- Never invent. No source, no fact: write
[GAP-nnn]inline and add the row togaps.md(file, what's missing, owner, exact question, blocking yes/no). - Every fact carries its source id; inferred facts are marked
(inferred, S3). - Constraints are written as
[MUST]/[MUST NOT]with something measurable; if breaking it has no named consequence, it is not a constraint — move it to the PRD. - Rules that must never break (D03–D03d) become invariants in
domain.md:[INV-nnn], the rule,class: <class>and the source. Ask how it breaks to pick the class; two ways → two invariants. Ids never change or get reused. Money, one-time side effects, limits and other people's data are where the worst bugs live: don't close the domain block without asking. - No vague phrases ("robust", "user-friendly", "best practices"…) —
python3 .keelokit/bin/doctor.pyrejects them. - No secrets, no personal data of real people; accounts are named by owner, never with values.
4. Keep the profile true
If the project has .keelokit/profile.toml, update its traits from what the context now says:
personal-data when it stores anything that identifies a person, payments when money moves,
i18n for more than one language, developer-facing when its users are developers. A new
feature that needs something new (an API, a database, a mobile app, hosting) adds it too — and
say so to the user: it changes which rules apply and what the bug bash hunts.
5. Check and hand back
- If the project has
.keelokit/bin/doctor.py, run it and fix everydocs/contexterror. - For a deep review, delegate to the
context-auditoragent (it didn't write the context, so it catches invention and vagueness the author misses). - Report: sources used, questions asked, gaps open by owner (blocking first), and the questions to ask next.
Resumable: the files on disk are the state. On a new session, read docs/context/ and
gaps.md, say what is covered and continue with the open gaps.
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.

