ramco-brain

Locates and inventories information inside the ramco-erp-brain data repo. Use for questions like "where is X stored", "what data do we have about component/module Y", "find the SPs/tables/services for activity Z", or to verify whether an input category (A01–C16) is actually present locally.

Onboard NEW journeys into the journey_v1 corpus (the executor-ready format-v4 journey brain) so they inherit every accumulated quality lane — JC waves, adversarial-verify fixes, Ramco review classes — by running the standing idempotent pipeline. Use this skill whenever the user asks to "add a journey", "update the brain with journeys", "onboard a new activity/module journey", "extend journey_v1", or when new activities/components need journey coverage. Canonical runbook: journey_v1/_build/NEW_JOURNEY_PROCESS.md (this skill is its executable form). Everything is idempotent over the whole corpus — re-running any step is always safe.

Carefully evaluate and grade the Ramco ERP "Brain" chatbot's answers against the ground-truth question bank (BrainEvals-Layer1). Use this skill WHENEVER you are scoring, grading, judging, benchmarking, or running an evaluation of Brain answers — or when the user says "eval the brain", "score these answers", "run the evals", "grade against ground truth", "judge as correct/hallucination", or asks how well the Brain answers. Two rules define careful grading here and this skill enforces them: (1) drop two-word and otherwise low-quality/malformed questions from the eval set — they are not real questions; (2) an answer that is MORE comprehensive than the ground truth (a superset) is CORRECT, not a hallucination — a hallucination is ONLY inventing facts that are not there. Judge on informational equivalence of the important points, semantically, in English.

Generate a deep, gold-quality "journey" JSON for a single Ramco ERP activity (e.g. Create Direct PO, Edit & Approve PO, Maintain Item Basic Information, Generate Supplier Rating, Print PO Register, Purchase Hub) from the Ramco data drop — by parsing the component's Service_details CSV, the stored procedures, the screen objects, and the integration cascade. Use whenever the task is to build, extend, fix, or validate a Ramco activity journey to the depth of the gold po_create_direct reference (14 screens / 197 slots / 146 tools / 493 rules), or to map an activity's screens, slots, SP chains, discriminators, rules, and cross-component (TCAL/VAT/workflow/budget) dependencies.

Safely ADD new knowledge to the Ramco ERP "Brain" (the LLM-wiki / vectorless-RAG knowledge system) as a strictly additive, non-regressing positive delta. Use this skill WHENEVER you are extending, reindexing, updating, or ingesting into the Brain — a new source file or artefact class, a new cross-connection/edge between components, or a new instruction in the chatbot's system prompt or tool set. Also use it whenever the user says "reindex", "add this to the brain", "ingest", "extend the brain", "add a parser/generator", "wire up a new source", or asks to update brain content without breaking what is already there. The governing rule is: the CURRENT Brain is pre-eminent — nothing is ever deleted, and earlier answers must never regress.

Onboard a batch of NEW or CHANGED Ramco source files into the Brain — the end-to-end playbook that runs WHENEVER a `git pull` of `ramco-erp-brain/` lands new/changed files, a new folder/BPC/artefact-class appears, or the user says "sync the brain", "a git pull brought new data", "ingest the delta", "new files were added to the source, update the brain", or "the source changed — reflect it". This is the DELTA ORCHESTRATOR: it detects the change set, documents it, classifies every path, and routes each class through the `reindex-brain` skill (§A/§B/§C), then proves no regression with the `eval-brain` skill before anything becomes visible. It composes those two skills — it does not replace them. Governing rule (inherited): every change is a strictly additive, non-regressing positive delta; the current Brain is pre-eminent; nothing is deleted, no prior answer regresses.