Common questions.
What is ccmd, in one sentence?
ccmd is a free polyglot analyzer for agent-configuration files (CLAUDE.md, AGENTS.md, .cursorrules, .grokrules) that scores the file you already wrote and points to the lines silently bloating your context.
How is this different from a CLAUDE.md generator like Genesis Claw or codewithclaude.net?
Generators emit a starter file when you pick a stack from a form. ccmd reads the file you already have, scores it, and tells you what to cut. They are complementary: generate first, then audit when the file has grown past 2k tokens.
How is this different from token-cost CLIs like ccusage or claude-meter?
Cost CLIs read local JSONL logs and tell you what you spent. They never look at CLAUDE.md. ccmd reads CLAUDE.md and tells you why you spent it. We integrate with both on the paid tier: paste the ccusage JSONL and we attribute spend to specific rules.
How is this different from Anthropic's claude-code-setup plugin?
claude-code-setup is read-only scaffolding that recommends 1-2 hooks/skills for a brand-new project. It does not score CLAUDE.md, does not estimate token cost, does not monitor drift, and does not recommend non-Anthropic skills. ccmd does all four. Anthropic structurally cannot ship a paid analyzer; the marketplace incentive is wrong for them.
Why not just ask Claude to optimize my CLAUDE.md?
Claude pasting its own file back at itself is a 200-token tip. ccmd measures: which lines fire on real session traces, which never do, which contradict the system prompt, which cost the most per useful output token. Claude cannot count its own tokens reliably and cannot see your billing.
Does it work with Codex / GPT-5.5?
Yes. AGENTS.md is the OpenAI Codex equivalent of CLAUDE.md; ccmd parses both with the same rubric. The Karpathy-12 rubric is model-agnostic.
Does it work with Cursor?
Yes. .cursorrules (single-file) and .cursor/rules/*.mdc (multi-file) are both supported. We treat the rules as flat input regardless of file shape.
Does it work with xAI Grok Build?
Yes. Grok Build CLI shipped May 14 2026 with a rules format very close to skills/CLAUDE.md. ccmd was polyglot from day one specifically to cover Grok Build alongside the other three.
Does it work with GitHub Copilot instructions?
Partial. .github/copilot-instructions.md follows a similar shape; ccmd will score it as 'unknown' for now but the same lints apply. Full Copilot support is on the 60-day roadmap.
Does my CLAUDE.md leave my browser when I use the free analyzer?
No. The free tier is a pure-frontend tool. Your file never hits a server. You can disconnect your wifi and the analyzer still works.
What does the paid tier send to your servers?
Two things, both opt-in: (1) the parsed structure of your CLAUDE.md (rule list, no surrounding code); (2) aggregate metadata from your Claude Code session logs (which rules fired, how many tokens, no prompt text and no output text). We never read your code, never log your prompts.
Are you SOC 2?
In progress, not certified. Reach out if you have a procurement gate; we will share the current control mappings.
What is the June 15 2026 Agent SDK billing change?
Starting 2026-06-15 Anthropic splits programmatic Claude Code usage (Agent SDK, claude -p, GitHub Actions, third-party agents like OpenClaw and Conductor) into a separate monthly credit pool at full API rates. Pro $20 / Max 5x $100 / Max 20x $200, no rollover. Interactive use (Claude.ai chat, Claude Code TUI, Cowork) stays on the existing subscription pool.
How does that affect my CLAUDE.md?
Programmatic agent runs (cron jobs, GitHub Actions, autonomous loops via /goal) now pay full API rates against a fixed credit pool. A bloated CLAUDE.md that re-reads the same files every turn burns programmatic credits 5-10x faster than it burns interactive subscription quota. ccmd flags those patterns specifically.
Can the whole team use one ccmd account?
On the team tier, yes. Up to 10 seats share one workspace with per-engineer cost attribution, PR diff comments on every CLAUDE.md change, and a shared dashboard. Need more seats? Talk to us.
Can we self-host the paid tier?
We will build self-hosted enterprise when the first customer asks; we are not building it speculatively. The free analyzer is already self-hosted because it lives in your browser.