Rig Doctor
How heavy and how safe is your agent setup? Get your context tax (tokens loaded every turn, by source), a guardrail score out of 5, your archetype and percentile against 6,413 real public rigs, and concrete fixes with ready-to-paste guardrail snippets. Same method as every rig page. Free.
We fetch only agent config at standard paths (CLAUDE.md, AGENTS.md, .claude/settings.json, .mcp.json, skill/agent frontmatter) from raw.githubusercontent.com, with size caps and timeouts. Env values, headers and anything secret-looking are redacted before analysis; nothing is stored. Repos with an .agentrigs-ignore file are skipped.
Save your rig and re-check it weekly
Rigs drift: a new MCP here, a skill pack there, and your context tax doubles. Pro supporters get Rig Doctor watch (weekly re-check of your repo, rolling out to supporters first), a supporter badge and the weekly data export, and keep this tool free for everyone.
$5 / month $39 / yearGet one guardrail tip a week
The Rig Report: rising MCPs and skills, standout new rigs, one stat and one guardrail tip per week. Free.
No spam, no sharing. Read past issues · privacy
How the Doctor scores
- Context tax = always-loaded instructions (characters ÷ 4 of your primary rules file plus
@imports) + skill and subagent metadata (name + description ÷ 4, +8 each) + an estimate of each MCP server's tool schemas (measured sizes for ~30 popular servers, 2,500 tokens for unknown ones). Index median: 2.2k. Details. - Guardrails (one point each): blocks destructive commands · protects secrets · PreToolUse hook · no YOLO mode · sandbox or ask-first rules. Details.
- Repo mode reads root-level agent files (plus
@imports, skill and subagent frontmatter). Rig pages also add nested CLAUDE.md files from subfolders, so they can show a higher instructions number for the same repo. - Percentiles compare you with every indexed rig: "leaner than" counts rigs with a higher tax, "safer than" rigs with a lower guardrail score.