domain-config-advisor
Skill8 starsChanged 4 days agoMITView on GitHub
Read business context (a description, website copy, sample transcripts/documents, or an interview with the user) and design, draft, and validate an imi domain schema (config/domains/*.yaml). Use when the user wants to adapt imi to their business, asks "which domain should I use", wants new entity types, or extraction keeps producing the wrong entities.
- Reads credentials
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- imi-onboardingSkill
- briefSkill
- constitution-reviewSkill
- memory-wrapSkill
- what-changedSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
Use whatever is available, in preference order:
- Very specific: line 25
If an imi instance is already running with data: list_entities per type and GET /api/type-registry — provisional types are unmet demand: …
- Distinctive: line 45
3–6 entity types: Every type dilutes extraction attention.
- Very specific: line 57
Add 2–5 intelligence_patterns (keyed dict.
- Very specific: line 65
Write config/domains/<id>.yaml.
≈ “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
Domain Config Advisor
You are designing the type system for a knowledge engine. The output is one YAML file that
drives extraction, graph validation, LLM analysis patterns, and UI labels. Read
docs/customization/domain-schemas.md for the full format; the authoritative validator is
app/model_schemas/domain_config.py (beware the divergent legacy model at
app/models/domain/config.py).
Step 1 — Gather business context
Use whatever is available, in preference order:
- Documents the user provides (about page, service descriptions, a sample call transcript, an org chart).
- If an imi instance is already running with data:
list_entitiesper type andGET /api/type-registry— provisional types are unmet demand: they're what users/agents tried to create that the schema didn't cover. - A short interview. The five questions that matter:
- Who do you sell to / serve, and what do you call them? (client? account? member? patient?)
- What unit of work do you track? (project? engagement? campaign? case? deal?)
- Which people matter and in what roles? (stakeholders? team? both?)
- What recurring risks or opportunities should call transcripts surface?
- What statuses/stages do you actually use today? (these become enums)
Step 2 — Choose base: existing, extend, or new
Compare against the six shipped schemas (config/domains/*.yaml): consulting_firm, b2b_saas,
agency, solo_consulting, member_network, personal_crm. If one covers ~80%, recommend it (or a
copy with small edits) over a new schema. Say which and why.
Step 3 — Design the entity model
Rules of thumb (justify deviations):
- 3–6 entity types. Every type dilutes extraction attention. "Would the user want a dashboard listing all X?" — if no, it's an attribute or a signal, not an entity.
- Things people say in meetings need
ner_labels(PERSON/PERfor humans,ORG/ORGANIZATIONfor companies). Addner_excludefor domain jargon that NER mistakes for organizations (standards bodies, acronyms, product names) — seesolo_consulting.yamlfor a real example. descriptionfields are read by the extraction LLM — write them as instructions ("Client organizations the practice advises"), not documentation.- Enums beat free strings for anything filterable (status, stage, tier, stance).
- Relationships: model the 2–4 traversals the user will actually ask about. Omit
inverse_nameunless a paired edge is needed — if set, the target entity MUST declare the reciprocal relationship or startup fails (validate_inverse_names). - Add 2–5
intelligence_patterns(keyed dict;pattern_type∈ risk/opportunity/escalation/ commitment/decision/insight) written as natural-language descriptions of situations, not rules. - Include a
ui:block — renaming "Entities" to the user's own vocabulary is cheap and makes the product feel native.
Step 4 — Draft and validate
- Write
config/domains/<id>.yaml.idmust be snake_case and equal the file stem;id,namerequired at domain level;name,description,pluralrequired per entity;enum:list required for enum attributes. - Validate mechanically before presenting:
(Outside docker, run the same with the repo on PYTHONPATH. A running instance also validates at boot — errors name the failing field.)docker exec imi-dev python -c " from pathlib import Path import yaml from app.model_schemas.domain_config import DomainConfiguration data = yaml.safe_load(Path('config/domains/<id>.yaml').read_text()) cfg = DomainConfiguration(**data.get('domain', data)) print('OK:', cfg.id, '-', len(cfg.entities), 'entities') " - Activate: set
ACTIVE_DOMAIN=<id>in.env, restart the app.
Step 5 — Prove it with a real ingest
Take a genuine sample document/transcript from the user (or synthesize a realistic one from
their context), post it to /api/ingest, and review the delta report and extracted entities/
signals with the user. Iterate on description and ner_exclude until extraction looks
right — one round of this is worth more than any amount of schema theorizing.
Deliverable
Present: (1) the recommendation (existing/extended/new + why), (2) the YAML, (3) validation
output, (4) results of the test ingest, (5) what was deliberately left out and the signal for
adding it later (watch /api/type-registry for provisional types).
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.

