| name | brain-ops |
| description | Brain-first local memory workflow for Codex. Use to search, update, cite, or cross-link the local brain before outside research or after new user signal. |
Brain Ops
Use this skill when the user wants memory-grounded work and the local brain/ corpus should influence the response or capture new context.
This port is adapted from garrytan/gbrain at commit 0b47afbf402a4e27a648bb9d131ce584461461ea.
Important Adaptation
Upstream GBrain treats brain ops as an ambient behavior layer. In this repo, the equivalent is an explicit workflow backed by local file operations:
python3 scripts/brain_search.py "<query>"
python3 scripts/brain_put.py ...
python3 scripts/brain_link.py ...
python3 scripts/brain_citations.py
Workflow
- Check the local brain first:
- search for the entity, topic, or project
- read the top matching pages before doing outside research
- If the user shared new information, write it back:
- create or update the right page under
brain/
- add a dated timeline entry
- Maintain graph hygiene:
- add backlinks for referenced people, companies, meetings, or concepts
- keep inline
[Source: ...] citations on any new fact
- Answer or continue the task using the retrieved context.
Current Upstream Coverage
Enforce the brain-first loop on inbound and outbound work: read relevant context, enrich or write durable signals, update structured relationships, then answer from refreshed context. Every new durable page must backlink to its source or parent context; cross-source answers must identify which source supports each claim.
Guardrails
- Do not claim this is always-on ambient capture; run it deliberately.
- Do not use external research before checking the local brain when the task is memory-sensitive.
- Do not invent citations or backlinks just to satisfy formatting.