What the brain holds and how it is kept — the agent's persistent memory of what is true. Load before restructuring or distilling it, or when unsure whether something belongs in the brain, a project's memory, or the notes.
Scanned 9/3/2026
Install to Claude Code
npx -y skills add w-is-h/agentslate --skill slate-brain --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Slate Brain?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/w-is-h-slate-brain)More formats (shields.io, HTML) on the badges page.
---
name: slate-brain
description: "What the brain holds and how it is kept — the agent's persistent memory of what is true. Load before restructuring or distilling it, or when unsure whether something belongs in the brain, a project's memory, or the notes."
---
# The brain
The brain is *what's true* — objective state: facts about the user, the world, the projects, where the work stands. Objective only; subjective texture belongs in the notes.
**Routing**: the brain holds only the general and high-level; anything project-specific, machine-local, or more granular goes down into that project's memory page (granular is fine down there) — the project map is `memory_list`, generated from the pages, never kept by hand. When in doubt, down, not up.
**Structure**: the user material in three sections — the record (bio, history), reading them (how they communicate, so their messages parse right), what lands (general preferences that shape responses) — plus the relationship. The user sections fill from a session read as a whole — where they were clearly pleased or annoyed, how they phrased things, what patterns emerged — pattern-level evidence only, hedged while tentative ("one session, tentative"); a single stray detail is page material, and domain taste (code style, frontend) goes to a skill of its own.
**Mechanics**: nothing here freezes — wet ink all the way down: correct, restructure, delete freely; if something stops being true, change it. Prefer `brain_edit`; `brain_set` only with content you wrote or read first.
The cap (5,000 chars by default, `brain_limit` in config.yaml) is hard: a refused write is the signal to distill — cut by at least 10%, up to half where garbage has accumulated; never just enough for the new line, never spill into a page.
**What doesn't belong**: subjective or identity material (→ notes), instructions and tool descriptions (→ your rules files), how the plumbing works (→ a README), dated "what I changed today" entries (→ notes). When the brain drifts toward any of those, move that content where it belongs.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!