Skip to main content

Documentation Index

Fetch the complete documentation index at: https://auto-sop.com/docs/llms.txt

Use this file to discover all available pages before exploring further.

​
Frequently Asked Questions

No. The hook shim adds less than 50ms of overhead per tool call. Captures use a double-fork technique — the capture process detaches immediately so Claude Code doesn’t wait for it. You won’t notice any difference in your workflow.
The free tier is $0 forever — 1 project, unlimited directives, all CLI commands included.LLM analysis (the learner) runs through your existing Claude Max subscription. auto-sop doesn’t call any paid API — it uses the same Claude instance you already have. The free plan includes cloud sync for 1 project (stats and dashboard access). The Pro plan (12/moor12/mo or 99/yr) adds unlimited projects, directive sync, and priority support.
You have multiple recovery options:
  • auto-sop revert — Restores CLAUDE.md from the most recent backup, removing the latest batch of directive changes
  • TTL pruning — Directives that don’t fire (aren’t useful) are automatically pruned over time
  • auto-sop repair — Fixes managed-section drift if you’ve manually edited the managed area
  • Manual edit — You can always edit CLAUDE.md directly; use auto-sop repair afterward to re-sync the hash
The learner requires 3+ occurrences of a pattern before writing a directive, and validates each candidate with LLM analysis. False positives are rare but recoverable.
auto-sop works with Claude Code CLI regardless of your editor. Whether you use VS Code, Cursor, Neovim, or a plain terminal, auto-sop hooks into the Claude Code process — not your editor.If you run Claude Code through your terminal or an editor’s integrated terminal, auto-sop captures and learns normally.
  • Free plan: 1 active project at a time. To switch, uninstall from the current project and install in a new one.
  • Pro plan: Unlimited projects. Each project syncs independently with its own stats and directives.
Run auto-sop install --license YOUR_KEY in each project directory to bind it.
Yes. auto-sop supports macOS, Linux, and Windows. The scheduler uses launchd (macOS), systemd (Linux), or Task Scheduler (Windows) depending on your platform.
The learner requires 3 or more occurrences of a pattern before considering it a directive candidate. After detection, each candidate is validated with LLM analysis to confirm it represents a genuine recurring mistake — not a coincidence or intentional behavior.This threshold keeps directives high-quality and evidence-based.
The managed section is hash-protected. auto-sop detects manual edits and marks the section as drifted.If you need to make changes:
  1. Edit CLAUDE.md as needed
  2. Run auto-sop repair to re-sync the hash
Alternatively, add your own directives outside the managed section — auto-sop only manages the clearly marked <!-- auto-sop managed --> block.
The managed section is clearly delimited with comment markers. Everything outside those markers is yours to edit freely.
No. Never. auto-sop captures Claude Code tool calls and outcomes — not your source code. Even these captures stay local on your machine.The data that can leave your machine (opt-in via license key) is:
  • Aggregate statistics (tokens saved, errors prevented) — all plans
  • Project slug (directory name) — all plans
  • First 10 words of each directive — Pro only
  • Directive fire counts — Pro only
See the Privacy & Data Safety page for the complete breakdown.
Run the health check:
auto-sop doctor
This verifies hooks, scheduler, managed section, and overall installation health. You can also check recent activity:
auto-sop status     # Installation status
auto-sop recent     # Recent captures
auto-sop stats      # Directive fire metrics