Spool.

Spool ships a /spool skill that any ACP-compatible agent can call mid-conversation. The agent searches your local session library and pulls matching fragments back into its context — no copy-paste, no cloud round-trip.

The /spool skill

The skill lives in skills/spool/SKILL.md in the repo. To use it from Claude Code, copy that file into your project's .claude/skills/spool/SKILL.md (or your global ~/.claude/skills/spool/SKILL.md).

The skill requires the spool CLI to be on PATH. Install it once:

bash
npm install -g @spool-lab/cli

Example

Inside a Claude Code conversation:

> /spool auth middleware refresh token rotation

Claude invokes the skill, runs spool search against your local index, and presents the top matches with source attribution (Claude / Codex / Gemini) and session UUIDs. You can ask Claude to load any of them with spool show <uuid> for full context.

How it works

  1. The agent invokes /spool <query>.
  2. The skill shells out to spool search "<query>" --json --limit 5.
  3. Matching fragments come back with source, project, startedAt, snippet, and uuid.
  4. The agent presents the results and offers to load any session in full.

Inference and search both happen on your machine. Spool's status bar always shows via ACP · local while AI mode runs — the same trust signal applies when an agent calls the skill.

Beyond Claude Code

Any agent that can call a CLI tool can use Spool the same way. The CLI's --json flag gives you a stable shape to parse; see CLI reference for the full surface.