analyze-har

analyze-har

Turn a HAR file into a structured map of the API surface it captured. This is the zero-install entry point — any browser can produce a HAR.

Inputs

• Path to a .har file. If the user pastes a URL or refers to a tab instead, redirect them to observe-tab (with claude-in-chrome) or capture-via-proxy (with mitmproxy).
• Optional: host filter — useful when the HAR contains noise from third-party trackers, fonts, analytics.

What to produce

Write under ${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/browser-data-capture/data/<run-id>/:

• endpoints.json — one entry per unique (method, host, path-template) tuple, with:
  • sample request URL, query params seen (with example values), request headers of interest (auth, content-type, custom x-*), request body shape if JSON
  • response status codes seen, response content-type, response body shape if JSON
  • count of times the endpoint appeared in the HAR
• schemas/<endpoint-id>.request.json and .response.json — inferred JSON Schema for bodies, when JSON
• summary.md — human-readable overview: hosts contacted, endpoint count per host, auth scheme(s) detected, notable patterns (REST vs GraphQL vs RPC, pagination style, ID formats)

Method

1. Parse the HAR (.log.entries[]).
2. Group entries by host. Drop static-asset CDNs (fonts, images, analytics) unless the user asked to keep everything — list what was dropped in the summary.
3. For each remaining entry:
   • Extract method, url, parse out path and query.
   • Templatize the path: replace numeric IDs and UUID-like segments with {id} / {uuid} placeholders so /users/123/posts and /users/456/posts collapse to one endpoint.
   • Capture request headers of interest (authorization presence only, content-type, x-*).
   • If request body is JSON, infer schema (use genson).
   • Same for response body.
4. Detect auth scheme: Bearer token, cookie session, API key in header, API key in query, none.
5. Detect API style: GraphQL (POST to a single /graphql endpoint with query field), REST (resource-shaped paths), RPC (verb-shaped paths).
6. Write the outputs.

Privacy

Never write raw cookie values, raw bearer tokens, or raw request/response bodies that contain auth secrets to the outputs. Replace with <redacted> and note presence. Users hand HAR files over expecting analysis, not a permanent copy of their session secrets.

Hand-off

When done, point the user at summary.md and offer:

• create-domain-map — produce a per-domain map document for version control.
• generate-openapi — produce a draft OpenAPI 3.1 spec.