Ontology Assembly - Playbook for Playbooks

Description

Map the Extraction Worksheet onto the Mimir schema. Wire everything together into a complete, executable playbook. Before release, intern-test: a capable person with no domain context must be able to start from an activity packet.

Activities in this Phase

Order	Name	Actions
Assemble Playbook
#1	Map goals to hierarchy
#2	Group activities into workflows	Read the 'AAP-2 Group activities into workflows' activity from playbook 'Playbook for Playbooks' (workflow 'Assemble Playbook'). Follow its guidance when assisting me. Cluster activities into coherent sequences that share a defined output. Assign each workflow a name, 3-letter abbreviation, and description. Grouping heuristics: - Pipeline stage: what triggers the next group? (data prep → model → evaluation is a natural boundary) - Shared artifact: activities that produce and consume the same intermediate artifact belong together - Role boundary: if a different skill or team takes over, that is often a workflow boundary - Size check: prefer 3–7 activities per workflow — fewer suggests over-splitting; more suggests under-splitting
#3	Group workflows into phases
#4	Wire predecessors
#5	Link agents, skills, artifacts, rules
#6	Validate coverage
#7	Create in Mimir	Read the 'AAP-7 Create in Mimir' activity from playbook 'Playbook for Playbooks' (workflow 'Assemble Playbook'). Follow its guidance when assisting me. Build the playbook in Mimir using the MCP tools. Create phases first, then workflows, then activities in order. Wire all links: agents, skills, artifacts, rules, and predecessors. Verify the final structure matches the validated worksheet. This is not the release gate — next is Run intern test.
#8	Run intern test	Read the 'AAP-8 Run intern test' activity from playbook 'Playbook for Playbooks' (workflow 'Assemble Playbook'). Follow its guidance when assisting me. Hand the assembled Mimir playbook to a capable intern or coding agent who has never seen this domain. Give them one activity at a time: name, guidance, and that activity's linked skills, artifacts, rules, and agent. No briefing, no extra files, no Extraction Worksheet. They must answer: from this packet alone, can I tell what needs to be done, or at least find pointers to the pieces I need to put together? Sample at least one activity per workflow, including the hardest assembly step — not only the easy ones. Record pass/fail and the intern's exact stuck-point in Intern Test Report. Pass: they can start without asking what the playbook is about, where inputs live, or which other work they must invent. Fail: they need tribal knowledge. Fix the activity (guidance and links), then re-test that activity. Do not release until the sample passes.