company-brain-demo

Applies and verifies the brain repo's docker-compose GBrain demo stack with persona-scoped access control.

Updated Dec 7, 2025
One-click install
npx skills add https://github.com/harlanljones/dotfiles --skill company-brain-demo-harlanljones
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: company-brain-demo
Source: https://github.com/harlanljones/dotfiles/tree/main/dot_hermes/skills/devops/company-brain-demo
Command: npx skills add https://github.com/harlanljones/dotfiles --skill company-brain-demo-harlanljones

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Re-applying the brain repo's v2 demo stack involves a fragile sequence of secret-store setup, recipe application, corpus git initialization, and smoke verification, where small mistakes (stale servers, concurrent applies, unsigned commits) silently wipe the stack or revert credentials. ## Core Features & Use Cases - Guided demo apply procedure: Walks through secrets.ts store setup, apply-recipe.ts execution, and serve-ui.sh restart in the correct order with fail-closed gates. - Pitfall catalog: Documents concrete failure modes such as git signing breaking corpus commits, concurrent applies overwriting .sandbox-credentials.json, and semantic retrieval invalidating denied probes. - LLM provider wiring: Explains how GBrain reads provider keys from Postgres config, the openrouter:<vendor>/<model> id format, and includes a scored model decision matrix in references/model-choice.md. - Use Case: After editing demo-v2.recipe.yaml, follow the skill to re-apply the recipe, restart serve-ui.sh, and confirm the per-persona smoke suite passes before presenting the demo. ## Quick Start Ask the assistant to re-apply the v2 demo recipe in ~/dev/brain and verify the smoke suite passes before launching the demo UI.

Frequently Asked Questions about company-brain-demo

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I re-apply the brain demo v2 recipe?▼

Run bun run scripts/apply-recipe.ts demo-v2.recipe.yaml after populating the cloud/local/.secrets store via secrets.ts. It rebuilds the stack with down -v, syncs corpus buckets, registers sources and OAuth clients, and runs the smoke suite. Restart ./serve-ui.sh afterward.

Why does the demo stack keep getting wiped or credentials revert?▼

Concurrent apply-recipe runs overwrite .sandbox-credentials.json and rewrite the stack via down -v. Also check for the legacy sandbox.test.ts suite, which ran docker compose down -v on every bun test until gated behind CB_LEGACY_SANDBOX=1.

Why are corpus bucket files invisible to GBrain sync?▼

GBrain's walker reads sources through git objects, so every corpus bucket must be a git repo with committed files. Host git signing or Conventional-Commit hooks can silently block commits; use commit.gpgsign false and git commit --no-verify.

How does GBrain configure the LLM provider and model?▼

GBrain reads provider keys from its own Postgres config, not environment variables, so entrypoint.sh must re-assert models.default, models.tier.deep, and the provider key on every container start. Model ids use the provider:model form, e.g. openrouter:anthropic/claude-haiku-4.5.

Why do denied probes fail in the smoke suite?▼

Semantic retrieval returns top-k matches for almost any query, so denied probes need vocabulary verified absent from the persona's allowed buckets. Also check the scorer in scripts/smoke.ts, since honest refusals listing allowed pages can trip naive keyword-leak checks.

Can the Drive collector access shared drives?▼

Yes, but only via the scoped path: recipe connectors.google_drive.folders flows through CB_DRIVE_FOLDERS into LiveDriveClient, which uses drives.list and files.list with corpora=drive. The plain /drive/v3/files listing never returns shared-drive files.