onboard-contour
Skill3 starsChanged 4 months agoMITView on GitHub
Use for the first-session Block A onboarding contour: a short ladder that frames the company, starting area, flow, source of truth, roles, and success metric before source setup.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- business-ontology-packageSkill
- build-brainSkill
- business-ontologySkill
- connect-sourceSkill
- daily-ingestSkill
- decide-like-moduleSkill
- drift-flagSkill
- drift-sweepSkill
- extract-from-inputSkill
- grill-gapsSkill
- interaction-contractSkill
- interpretSkill
- meeting-recorderSkill
- meeting-transcript-ingestSkill
- mine-materialsSkill
- package-updateSkill
- promote-digestSkill
- propose-changeSkill
- show-modelSkill
- synthesize-digestSkill
- system-analysisSkill
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 8 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Use this skill at the start of the first session.
- Distinctive: line 18
Use this skill when:
- Very specific: line 24
Do not use it for a deep modeling workshop.
- Distinctive: line 30
Start by inviting voice input:
- Distinctive: line 37
Ask one question at a time:
- Very specific: line 57
Before sending each unanswered setup question, record a human_request with kind=setup.
- Very specific: line 61
If the company model language is unanswered, keep it as pending-owner-selection, leave the human_request open, and do not mark onboarding …
- Distinctive: line 67
Ask one question at a time.
- Distinctive: line 72
Do not ask for review owners during onboarding.
- Distinctive: line 74
Do not mark anything accepted.
- Distinctive: line 75
Do not infer company model language from chat language.
≈ “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
Onboard contour
Purpose
Use this skill at the start of the first session. The goal is a usable contour, not a complete ontology. The owner spends 10 minutes giving enough shape for the agent to start reading sources. The contour includes the company model language: the language used for human-facing model text. It is not inferred from chat language.
When to use
Use this skill when:
- a new resident agent starts with an owner;
- the current model has no agreed business boundary;
- the owner wants to reset the starting contour.
Do not use it for a deep modeling workshop. If the owner wants to model a
process live for 60-90 minutes, use the capture loop in the primary
business-ontology skill.
Procedure
Start by inviting voice input:
You can answer by voice if it is easier. I can work from the transcript, and
voice usually carries more context. I will not store raw audio in the model.Ask one question at a time:
- What does the company do, in one paragraph?
- What do you produce or sell, and to whom?
- What directions, businesses, or product lines are inside it?
- What hurts most right now?
- Recommend the starting area yourself: "I will start with
<area>because <pain/source>. OK?" - What mainly flows through this area?
- Where does the truth about that flow live?
- Who are the key roles in this area?
- Which metric says this area is working well?
- Which language should I use for the company model text? Recommend the language in which the owner and team make decisions. Technical ids stay stable and language-independent.
The recommendation in step 5 is the agent's job. Use answers 3 and 4 plus any available source readiness. Do not ask the owner to choose from a blank slate when you can make a defensible recommendation.
Before sending each unanswered setup question, record a human_request with
kind=setup. When the owner answers, close that request and continue the
ladder. Questions answered in the same incoming message may be recorded and
closed immediately so the ledger still explains why no setup ask remains open.
If the company model language is unanswered, keep it as
pending-owner-selection, leave the human_request open, and do not mark
onboarding complete.
Rules
- Ask one question at a time.
- Short answers are enough; do not force workshop-level detail.
- Stage candidate material as soon as an answer gives enough evidence.
- Mark unknowns as
unknownand continue. - Voice is welcome; treat transcripts as source content and redact by policy.
- Do not ask for review owners during onboarding. The owner is the starting reviewer until evidence shows otherwise.
- Do not mark anything accepted. Contour answers become candidate proposals.
- Do not infer company model language from chat language.
Output
Candidate material for:
- company or business card;
- selected starting area;
- primary flow object;
- source-of-truth hypothesis;
- key roles;
- success metric with formula
unknownwhen needed; - company model language or an open setup
human_requestif it is still pending; - open questions from pain points and unknowns.
Then hand off to connect-source for Block B.
What good looks like
The owner answers briefly. The agent records a skeleton, recommends one starting area, confirms it, and moves to sources. The chat stays plain; technical ids stay in staged artifacts.
Eval cases
Case 1 - owner answers everything in one voice note. What good looks like: the agent splits the transcript into the ladder fields, summarizes the candidate contour back in plain language, asks only for the single missing confirmation of the recommended starting area, and stages candidate material. It does not ask the whole ladder again.
Case 2 - owner does not know the source of truth.
What good looks like: the agent records source of truth as unknown, keeps the
metric formula unknown if necessary, and proceeds to source setup. It does not
invent a CRM or dashboard.
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.

