← ClaudeAtlas

openclaw-auditlisted

Trim an OpenClaw workspace's base system prompt. Measures what the always-injected workspace files (SOUL.md, AGENTS.md, IDENTITY.md, USER.md, MEMORY.md) cost per turn, then works through them section by section — cutting what the tools or codebase already say, moving durable conventions to Clawness ranked retrieval (.clawness/rules/) and one-line traps to .clawness/memory.md, and leaving load-bearing persona and identity where they are. Destructive and deliberate: run it when you have the session to spend.
fullymiddleaged/Clawness · ★ 4 · Code & Development · score 77
Install: claude install-skill fullymiddleaged/Clawness
# Trim an OpenClaw workspace prompt OpenClaw assembles its system prompt fresh for every agent turn. Alongside the fixed sections it manages (Tooling, Safety, Skills, Workspace, …), it injects the workspace's own Markdown files **in full, every turn**: `AGENTS.md`, `SOUL.md`, `IDENTITY.md`, `USER.md`, `BOOTSTRAP.md` (new workspaces only, self-deletes after first run), and `MEMORY.md` when present. That injected block is the one part of the per-turn cost the workspace owner controls — and the usual "landslide of noise per turn" complaint is it growing unchecked. OpenClaw caps it at 20,000 chars per file and 60,000 total and truncates past that with a terse notice, which is a worse outcome than trimming on purpose: you don't choose what gets cut. This skill is the **remedy**, the OpenClaw sibling of `/clawness:claude-md`. It is opt-in work: you invoking it is the consent, because a good trim is destructive and takes a while. **Say this before starting.** This reorganises files the user may care about a great deal (SOUL.md *is* the agent's personality), and it takes most of a session. Tell them: nothing is deleted until its replacement is verified, and they can stop after any single file. Then get an explicit go-ahead. > **Filenames and limits below are from OpenClaw's public docs, not a live install.** > Before you measure, confirm the workspace path and the file set actually present — see > the same caution in [openclaw/README.md](openclaw/README.md) ("Assumptions to verif