agentoscope

See every instruction your repo gives to AI coding agents — in one visual dashboard.

Website & live demo · npm

Modern repos accumulate agent instructions everywhere: AGENTS.md, CLAUDE.md, .cursor/rules/*.mdc, .github/copilot-instructions.md, .windsurf/rules, and a dozen more. agentoscope scans your repo for all of them and opens a beautiful, self-contained viewer so you can audit exactly what your agents are being told — and how much of their context window it costs.

npx agentoscope

That's it. It scans the current directory, starts a local server, and opens the viewer in your browser.

What it finds

| Provider | Files detected | | --- | --- | | AGENTS.md (universal: Codex, Cursor, Kimi Code, Amp, Jules, OpenCode…) | AGENTS.md at any depth | | Claude Code | CLAUDE.md, CLAUDE.local.md at any depth | | Cursor | .cursor/rules/**, .cursorrules | | Windsurf | .windsurf/rules/**, .windsurfrules | | GitHub Copilot | .github/copilot-instructions.md, .github/instructions/** | | Gemini CLI | GEMINI.md at any depth | | Cline | .cline/rules/**, .clinerules | | Roo Code | .roo/rules/**, .roorules | | Kilo Code | .kilocode/rules/**, .kilocoderules | | Continue | .continue/rules/**, .continuerules | | Aider | CONVENTIONS.md | | Goose | .goosehints | | Amazon Q Developer | .amazonq/rules/** | | Warp | WARP.md | | Junie (JetBrains) | .junie/guidelines.md | | Trae | .trae/rules/** | | Zed | .rules |

Why it's useful

• Token budget — every rule file is injected into some agent's context. See the estimated token cost per file and per provider at a glance.
• Duplication & drift — detects symlinks (CLAUDE.md → AGENTS.md) and identical-content duplicates so you know when rules are copy-pasted and can drift apart.
• Rule targeting — parses .mdc / .instructions.md frontmatter (globs, applyTo, alwaysApply) and shows it as badges, so you can see which rules attach where.
• Effective context — type any repo path (in the viewer or via --for) and see exactly which rules an agent receives when working on it, and what it costs.
• Lint your rules — agentoscope check catches stale globs that match nothing, oversized files, conflicting root instructions, and Cursor rules that can never auto-attach. Exits non-zero on errors, so it drops straight into CI.
• Scan any public repo — agentoscope.rafael.ltd/r/owner/repo renders a live report for any public GitHub repo, no install.
• One-click context export — "Copy all context" concatenates every rule file into a single blob for pasting into an LLM.
• Shareable report — export a single self-contained HTML file you can attach to a PR or wiki.

Usage

agentoscope [directory] [options]
agentoscope check [directory]      Lint agent rules (CI-friendly; exits 1 on errors)

-o, --output <file>  Write a self-contained HTML report and exit
    --json           Print the scan report as JSON (for scripting/CI)
    --for <path>     List the rule files that attach to a repo path and exit
    --no-open        Start the server without opening a browser
-p, --port <n>       Port for the local server (default: random)

In the viewer: / focuses the filter box, Esc clears it, and the top-bar path input shows the effective context for any file.

Examples

npx agentoscope                      # scan cwd, open viewer
npx agentoscope ../other-repo        # scan another directory
npx agentoscope -o report.html       # export a shareable HTML report
npx agentoscope --json | jq .stats   # pipe the report into scripts
npx agentoscope check                # lint rules (great in CI / pre-commit)
npx agentoscope --for src/api.ts     # which rules attach to this file?

In CI

- run: npx agentoscope@latest check   # fails the build on stale/broken rules

How it works

Zero-runtime-dependency CLI: the scanner walks your repo (skipping node_modules, build output, etc.), matches paths against a registry of known agent conventions, parses frontmatter, and embeds everything as JSON into a single HTML file with the viewer inlined. Nothing leaves your machine — the server binds to 127.0.0.1 only.

Development

npm install
npm run build       # bundle CLI + viewer into dist/
npm test            # vitest (scanner, report, viewer)
npm run typecheck

License

MIT