claude-mega-brain
OKF knowledge navigation. ALWAYS use this skill the moment a `<mega-brain>` block appears in session context — it means the…
Create or update a single OKF concept file from raw input. Use when the user wants to document a specific table, metric, API, service, or concept in the knowledge base — even if they say "add this to the okf", "documenta essa tabela", "cria um conceito pra isso", "adiciona no
$ npx -y skills add guhcostan/claude-mega-brain --skill ingest --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/ingestContext preview
The summary Claude sees to decide when to auto-load this skill.
Create or update a single OKF concept file from raw input. Use when the user wants to document a specific table, metric, API, service, or concept in the knowledge base — even if they say "add this to the okf", "documenta essa tabela", "cria um conceito pra isso", "adiciona no
name: ingest description: Create or update a single OKF concept file from raw input. Use when the user wants to document a specific table, metric, API, service, or concept in the knowledge base — even if they say "add this to the okf", "documenta essa tabela", "cria um conceito pra isso", "adiciona no knowledge base", "register this API", or pastes a schema/description and says to save it. If no OKF files exist yet, suggest /mega-brain:init first.
Turns raw input (schema dump, API doc, user description, table definition) into one OKF-formatted `.md` file committed to the knowledge base.
Look for existing OKF files: any `.md` with `type:` frontmatter, or `index.md` / `log.md` in the project. If none exist, stop and tell the user to run `/mega-brain:init` first.
If `.mega-brain.json` defines `dir`, save new concepts under that directory. Otherwise, choose a sensible location (`docs/`, project root, etc.) consistent with existing OKF files.
Check if a concept with the same name already exists. If so, ask whether to update it or create a new one.
One `.md` file per concept. Only `type` is required:
--- type: <concept type> title: <display name> description: <one sentence, plain language, no jargon> resource: <URL to the actual resource, if any> tags: [<tag1>, <tag2>] timestamp: <ISO 8601 date> --- <body: exact schema, formula, join patterns, examples, caveats>
| Type | Use for | |------|---------| | `BigQuery Table` | BQ tables with schema | | `BigQuery Dataset` | BQ datasets | | `Table` | Non-BQ database tables | | `Metric` | KPI or business metric with formula | | `API` | REST/GraphQL endpoints | | `Runbook` | Operational procedures | | `Concept` | Business or domain terminology | | `Service` | Internal microservices | | `Pipeline` | ETL or data pipelines |
Types are freeform — add your own.
1. Read the source material 2. Identify the concept type and canonical name 3. Write `description`: one sentence, plain language 4. Add `resource` URL if there's a live resource (BQ console URL, API endpoint, dashboard) 5. Write the body with exact values — preserve field names, types, and formulas as-is; do not generalize 6. Link to related concepts with `[Name](relative-path.md)` — this builds the knowledge graph 7. Save next to related OKF files (e.g. `docs/tables/<slug>.md` or `<config-dir>/tables/<slug>.md`) 8. Add one line to `index.md`: `- [Title](path) — description` 9. Append to `log.md`: `<date> — added <title>`
Loads the knowledge. Skips the search. 100% accuracy · 0 tool calls · −66% tokens vs Obsidian+MCP Real agentic sessions. Benchmark →
OKF knowledge navigation. ALWAYS use this skill the moment a `<mega-brain>` block appears in session context — it means the…
Initialize OKF knowledge base files in the current project. Use when the user wants to start documenting their project for…
Scan the project and migrate existing documentation into OKF format. Use when the user wants to populate the knowledge base from…