Agent instructions are scattered everywhere — AGENTS.md, CLAUDE.md, Cursor rules, Copilot instructions, and a dozen more conventions. One command finds them all and shows you exactly what your coding agents are being told, and what it costs in tokens.
agentoscope knows where each AI coding tool keeps its instructions — including nested files that only apply to part of your tree.
Agent rules are invisible context. Make them visible, measurable, and reviewable.
Every rule file rides along in some agent's context window. See the estimated token cost per file and per provider at a glance.
Spots symlinks (CLAUDE.md → AGENTS.md) and copy-pasted duplicates, so rules that drift apart get caught before they contradict each other.
Parses .mdc and .instructions.md frontmatter — globs, applyTo, alwaysApply — and shows exactly which files each rule attaches to.
Export a single self-contained HTML file. Attach it to a PR, a wiki page, or a design doc — no server, no dependencies, works offline.
"Copy all context" concatenates every rule file into one blob — perfect for pasting into a chat model that can't read your repo.
Zero runtime dependencies, binds to 127.0.0.1 only. Your code never leaves your machine.
No install needed — npx runs it straight from the registry. Requires Node 18+.
# scan the current repo, open the dashboard $ npx agentoscope # scan another directory $ npx agentoscope ../other-repo # export a shareable, self-contained report $ npx agentoscope -o report.html # machine-readable output for scripts and CI $ npx agentoscope --json | jq .stats
| -o, --output <file> | Write a self-contained HTML report and exit |
| --json | Print the scan report as JSON (files, providers, stats, frontmatter) |
| --no-open | Start the local server without opening a browser |
| -p, --port <n> | Pin the server port (default: random) |
| / · Esc | In the viewer: focus the filter box / clear it |
Proud of your agent setup? Add the badge next to your AGENTS.md: