mcpbeat

Gbrain Lite

cosmicstack-labs/gbrain-lite

Lightweight personal knowledge base — markdown + YAML frontmatter structured notes with full-text search and cross-referencing for AI agents

691 tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
365
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/cosmicstack-labs/mercury-agent-skills --skill gbrain-lite

The instruction itself

9 sections, as written by the author

GBrain Lite — Lightweight Personal Knowledge Base

> Organize books, people, concepts, and news items from conversations into a searchable, cross-referenced markdown knowledge base.

Workspace

  • Knowledge base directory: brain/
  • Subdirectories by type: books/ people/ meetings/ ideas/ articles/ projects/ news/
  • One entry = one .md file
  • News entries by date: news/YYYY-MM-DD/

Workflow

  • Trigger check — Book/person/concept/news mentioned in conversation → proactively check brain
  • Search — Full-text search in brain/ directory (ripgrep/grep)
  • Decide
  • Existing entry → reference it, update if needed
  • No entry + important → create immediately
  • No entry + uncertain → ask user
  • Create — Write brain/{type}/{slug}.md
  • Slug: lowercase/hyphenated
  • Must include YAML frontmatter (title, type, date, tags, summary)
  • Cross-reference — Update links field in related entries
  • Sync — Keep key entry index in agent memory

Rules

  • Don't dump long content directly into agent memory — prefer brain
  • Don't create entries without title and date
  • Don't skip the summary field — search and listing depend on it
  • Don't manually write updated — it auto-updates on save
  • News entries must use item_id as dedup key (format: github:owner/repo / hn:12345 / arxiv:url)
  • Don't do full overwrite updates on existing entries — use targeted patches

Validation

  • [ ] New entry has complete frontmatter (title, type, date, tags, summary)
  • [ ] Full-text search finds the new entry
  • [ ] News entries contain item_id + first_seen + star_history (GitHub entries)
  • [ ] Agent memory usage below 90%, otherwise trigger distillation to brain

Pitfalls

Brain vs Memory confusion

  • Symptom: Long-form content consumed all agent memory space
  • Root cause: Everything dumped into memory instead of brain
  • Fix: >500 chars → brain entry; memory only keeps index pointers. (Fixed: 2026-05-12)

Missing item_id on news entries

  • Symptom: Duplicate news entries accumulate, dedup impossible
  • Root cause: Agent skipped item_id field when creating news entries
  • Fix: Always include item_id for news entries, validated in creation workflow

References

  • references/garry-tan-gbrain-inspiration.md — GBrain system design reference

How to use it

Copy the folder

Take cosmicstack-labs/gbrain-lite from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.