Index of every page on this site: llms.txt
All skills
vercel-labs avatar

/derive-client

@5aad200 official
by Vercel Labsvercel-labs/agent-browser53k stars
3,714
Run it yourself

Reverse-engineer a website's internal API by recording browser traffic into a HAR file, then generate a standalone client or CLI that calls the endpoints directly, with no browser needed after the first recording. Use when asked to "derive a client", "build a CLI for <site>", "reverse engineer this site's API", "record network requests", "turn this site into an API", or when the same site will be automated repeatedly and direct HTTP calls would beat driving the browser every time.

Use this Skill: https://skilld.dev/gh/vercel-labs/agent-browser/derive-client

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈125 tokens for metadata: the name and description. ≈1.3k when used: this file.

Description uses 6.3% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈501 characters, or 6.3%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Derive an API client from a recorded session

Driving a browser is the right tool for the first visit and the wrong tool for the hundredth. This skill records a site's network traffic once while you use it, then turns the captured requests into a standalone client (script, CLI, or library) that talks to the site's internal API directly.

The recording alone contains everything needed: agent-browser embeds text response bodies (JSON/HTML/JS) in the HAR by default, so endpoint shapes can be studied offline after the browser is closed.

For a task the agent itself will repeat, agent-browser flow start and flow save do steps 1 to 4 in the core and flow run replays it without a browser. Use this skill when you need a standalone client outside agent-browser.

Workflow

1. Record     Start HAR capture, drive the flows you want in the client
2. Identify   Find the real API endpoints among the noise
3. Extract    Pull request shapes, response schemas, and auth material
4. Generate   Write the client, one function per flow
5. Verify     Call every endpoint for real before declaring done

1. Record

agent-browser network har start          # embeds text response bodies by default
# ... drive the site: search, open a detail page, paginate, etc. ...
agent-browser network har stop /tmp/site.har
  • Exercise every flow the client should support, and run each one at least twice with different inputs (two search terms, two detail pages). Diffing the recorded URLs reveals which parts are parameters.
  • If the site needs login, log in before starting the HAR so credentials don't land in the recording unnecessarily. The session cookies are exported separately in step 3.
  • --content all embeds binary bodies too (base64); --content none disables embedding. Per-body cap is 2 MB.

While the session is still open, agent-browser network requests and network request <id> give the same data interactively — but only the HAR survives navigation and browser close, so prefer it for anything multi-page.

2. Identify endpoints

Query the HAR with jq:

# All JSON API calls: method, URL, status
jq -r '.log.entries[]
  | select(.response.content.mimeType | test("json"))
  | "\(.request.method) \(.response.status) \(.request.url)"' /tmp/site.har

Ignore analytics and infrastructure noise: telemetry endpoints (/collect, /track, /beacon, /log), third-party domains (google-analytics, segment, sentry, datadog, intercom, hotjar), and static assets. The real API is usually first-party, JSON, and correlates with the actions you performed.

3. Extract shapes and auth

# Full detail for one endpoint: request headers, POST body, response body
jq '.log.entries[] | select(.request.url | test("api/search"))
  | {request: {method: .request.method, headers: .request.headers,
     postData: .request.postData.text},
     response: .response.content.text}' /tmp/site.har
  • Response schema: read .response.content.text — this is the real payload, use it to derive types.
  • Auth: compare request headers across endpoints. Look for authorization, cookie, x-csrf-token, x-api-key, and site-specific x-* headers. Replay only the ones that matter — test by omission in step 5.
  • Cookies: export the live session with agent-browser cookies get --json > cookies.json for the client to load at runtime. Never hardcode cookie values into generated source.

4. Generate the client

  • One function per recorded flow (search(query), getItem(id)), typed from the observed response bodies.
  • Auth material (cookies, bearer tokens) loads from a file or environment variable, with a clear error telling the user to re-run the browser login when it expires.
  • Reproduce the headers the API actually requires — some sites 403 without a matching user-agent, referer, or x-requested-with.
  • Keep pagination, sort, and filter parameters that appeared in the recorded query strings as function options.

5. Verify

Call every generated function against the live API and compare the response shape with the recording. Common failures:

Symptom Cause Fix
401/403 Expired or missing session Re-login via agent-browser, re-export cookies
403/419 on writes CSRF token is per-session or per-form Fetch the token endpoint first, or keep that flow browser-driven
Works then breaks Signed/expiring request params Fall back to the browser for that step; derive the rest
Different shape than HAR A/B tests or geo-dependent responses Re-record and treat the union as optional fields

Caveats

  • Internal APIs are unversioned and change without notice — keep the HAR so the client can be re-derived.
  • Respect the site's terms of service and rate limits; add delays for bulk fetching.
  • HAR files contain live session credentials (cookies, tokens, POST bodies). Treat them like secrets: keep them out of version control and delete them when done.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning14d3 checks · Risk SAFE
  • Gen Agent Trust Hub14d

    This skill provides a workflow for reverse-engineering web APIs by capturing and analyzing browser traffic. While it handles sensitive data like session cookies and generates code from external observations, it includes clear guidance on managing these secrets safely.

  • Socket14d

    No alerts

  • Snyk14d

    Risk: MEDIUM · 1 issue

Signed by skilld at 5aad200. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 1 hour ago.

Activeupdated 2 weeks ago
What it can do
Runs commands
All 2 allowed tools
Bash(agent-browser:*)Bash(npx agent-browser:*)

README badge

README badge for vercel-labs/agent-browser/derive-client