Brain ops

Skill Walliiee/agent-harness/skills/brain-ops

A drop-in self-improving memory + operations harness for AI agents (OpenClaw / Hermes / Claude Code / Codex). Layered memory, autonomous drift-fixing loop, evals, and one-command DR. Templates + scripts, no secrets.

Install

npx -y skills add Walliiee/agent-harness --skill brain-ops

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Structured gbrain write operations: create pages, write-back, update sources, citations, sync. Not for queries — use gbrain-query for reads. Not for default memory recall; use QMD first there.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.2 KB, 629 tokens by cl100k_base, as published. Nobody here has run it

Brain Ops

Use this skill for write operations against the user's structured gbrain knowledge base. For read/query operations, use the gbrain-query skill instead. For prior work, decisions, dates, people, preferences, or todos, start with qmd query unless the user specifically asks for gbrain.

Environment

Codex/OpenClaw may run with a synthetic home directory. Always target the real gbrain home explicitly:

GBRAIN_HOME=$HOME gbrain ...

The active brain is already configured at $HOME/.gbrain/config.json and uses Postgres + pgvector. Do not run bare gbrain init.

When to Use Brain Ops (writes)

The user asks to save, create, update, or write a brain page.
You need to register a source: gbrain sources register.
You need to sync or update structured knowledge.
You are writing durable structured knowledge to gbrain after the destination is clear.

When to Use Gbrain Query (reads) Instead

The user asks to look up, search, or query gbrain.
You need backlinks, timeline, or graph traversal for context.
The task is answer-only and does not need to write to gbrain.

When to Use QMD Instead

The user asks about prior work, preferences, decisions, todos, or dates.
You need fast recall across workspace memory/wiki/agent files.
The task is answer-only and does not need gbrain page/graph/timeline semantics.

If both are relevant, use QMD first for recall and gbrain second for structured source/page verification.

Write Flow

Gbrain writes are durable knowledge-base mutations. Only write when the user asked to save/create/update a brain page, or when an existing workflow clearly requires it.

Before writing:

Search for an existing page to avoid duplicates.
Confirm the correct source/workspace if multiple sources could fit.
Avoid PII/secrets. Use role-based identifiers where needed.
Preserve source attribution in the page body.

After writing, run the smallest relevant verification:

GBRAIN_HOME=$HOME gbrain get <slug>

If the write should be searchable immediately, run the existing sync path rather than inventing a new one:

GBRAIN_HOME=$HOME gbrain sync --source <id>

Safety

Never run gbrain init from Codex/OpenClaw synthetic HOME.
Never use gbrain as a reason to bypass QMD memory recall rules.
Never dump gbrain logs without redacting admin/access tokens.
Do not auto-enrich every inbound message. Surface or save only high-signal information that belongs in a durable brain page.
Prefer read-only commands until the target page and source are clear.

Report Format

For write work, report:

Updated: <source_id:slug>
Verified: <command/result>
Follow-up: <only if needed>