MADRank Sync

Report your measured AI usage to MADRank — the neutral usage layer for AI — without installing a CLI.

Works with Claude Code, Codex, OpenCode, Gemini CLI and DSH local logs on your machine.

Agent compatibility

SKILL.md + YAML frontmatter is the industry standard — these agents all consume the same skill file; only the install directory differs:

Agent	Format	Install location	Notes
Claude Code	SKILL.md + frontmatter	~/.claude/skills/madrank-sync/	verified
DeepSeek Harness (DSH)	SKILL.md + frontmatter	~/.dsh/skills/madrank-sync/ (or project .dsh/skills/)	verified
Codex CLI	no skill system	AGENTS.md instructions (skills/codex/AGENTS.md in this repo)	instructions variant, same collector
OpenClaw	SKILL.md + YAML frontmatter	per OpenClaw docs (docs.openclaw.ai/tools/skills); bundled + local overrides	format verified against official docs
WorkBuddy / CodeBuddy	SKILL.md + frontmatter	~/.codebuddy/skills/madrank-sync/ (or project .codebuddy/skills/)	format verified against official docs
ZCode	SKILL.md + frontmatter (name + description required, desc ≤1024)	Settings → Skills (per docs)	format verified; this skill description is short — passes
Trae (TraeCode)	SKILL.md + scripts/resources	per TraeCode docs (docs.trae.cn)	format verified against official docs

Three real prerequisites beyond the file format:

1. Runtime: the collector is a Node script (≥18.17; ≥22.5 only for the OpenCode SQLite source, which degrades gracefully). The agent must be able to run node.
2. Local data: the machine must have one of the supported CLI log dirs (~/.claude, ~/.codex, ~/.local/share/opencode, ~/.gemini, DSH store). No logs → the script honestly reports "not found" — it never fabricates.
3. Egress: the upload needs network access to madrank.ai.

The integrity model is host-agent independent: the script computes and uploads by itself, verifies its own sha256 pin, and never uploads without --yes — so the guarantee holds identically on every agent above.

Install

Claude Code (marketplace)

claude skills add qomob/madrank-sync madrank-sync

Manual (any agent, copy the folder to the target dir from the table above)

cp -r skills/madrank-sync ~/.claude/skills/madrank-sync     # Claude Code
cp -r skills/madrank-sync ~/.dsh/skills/madrank-sync        # DSH
cp -r skills/madrank-sync ~/.codebuddy/skills/madrank-sync  # WorkBuddy

Verify the pin before first sync:

shasum -a 256 skills/madrank-sync/scripts/collect.mjs
# must match skills/madrank-sync/scripts/collect.mjs.sha256

How it stays honest

• The script is the only component that computes, verifies, and uploads usage. An agent never derives numbers, never sees an alterable payload.
• A tampered collector refuses --sync (self-hash check against the pinned sha256). No explicit --yes means no upload.
• Per completed UTC day only; the current day is never transmitted.
• Shared installation id with the CLI/plugin (~/.madrank/usage/installation-id) — one machine, one node; server-side monotonic upsert prevents double counting.

Updating this skill

This folder is a distribution copy of the canonical collector built from @qomob/madrank-node. Refresh it:

node tools/update.mjs          # pulls the latest npm version
node tools/update.mjs 0.2.1    # or pin an exact version

Privacy

Uploads: anonymous id + completed UTC dates + per-model token counts. Never: prompts, responses, files, tool args, keys, or local paths. Opt out anytime via the CLI delete or the site.

MIT © qomob