version: "1.0.1" name: anakin-wire description: Use when a task targets a specific well-known website — extracting products, listings, prices, profiles, reviews or dashboard data from sites like Amazon, Walmart, LinkedIn, Airbnb or Zillow, or performing an interaction there such as submitting a form, adding to a cart, or posting content. Covers Anakin's Wire catalog of pre-built actions - wire_discover, wire_catalog, wire_read_action, wire_write_action, wire_identities, wire_login, wire_build, wire_build_status. Check this before hand-scraping any popular site.

Anakin Wire: pre-built site actions

Wire is a catalog of vetted, pre-built automation actions across hundreds of popular websites. An action returns clean structured data — not markdown you have to parse.

The single most important point

Wire is not login-only. Actions come in two kinds, and each has its own execution tool:

• READ actions extract data and change nothing — search listings, fetch a

category's products, get a product's price/specs/reviews, read a profile, pull dashboard metrics. Run with `wire_read_action`.

• WRITE actions change state — submit a form, add to a cart, post or send

content, update account settings. Run with `wire_write_action`.

Many read actions need no authentication at all. If you assume Wire is only for logins and skip it, you will hand-scrape sites that already have a one-call extractor. For any task on a recognizable site, try wire_discover first.

The loop

wire_discover (or wire_catalog)  →  check the action's type
                                      ├─ "read"  → wire_read_action
                                      └─ "write" → wire_write_action
                                            ↓ auth error only
                         wire_identities / wire_login → retry with credential_id

1. Find an action

• wire_discover — natural-language intent → ranked candidate actions. Pass q

(e.g. "top phones on walmart", "search airbnb listings in Lisbon"), and optionally limit (default 5). Returns each candidate's action_id, required and optional params, credit cost, and whether auth is needed.

• wire_catalog — browse instead of guess. No arguments lists every supported

site and its action count. Pass a slug ("walmart", "amazon", "linkedin") for that site's full action list with exact parameter schemas, each action's type (read/write), auth_mode (none/optional/required), credit cost, and login fields.

Use wire_discover when you know the intent, wire_catalog when you want to see everything a site can do.

2. Run it — pick the tool that matches the action's type

Both take action_id plus params matching that action's schema, and both poll the async job to completion for you. Confirm the action's `type` from discovery before choosing — passing a write action to wire_read_action (or the reverse) is an error, not a fallback.

• wire_read_action — for type: "read". Most need no auth.
• wire_write_action — for type: "write". Most do need auth.

Neither executes payments or transfers funds; such actions are refused.

3. Authenticate, only if the action demands it

If execution returns AUTH_REQUIRED, AUTH_EXPIRED, or FORBIDDEN, the error text tells you what to do. In short:

• wire_identities — list saved identities and their credentials. Each

credential's id is the credential_id you pass. Filter with catalog_id. Check the credential's status is `active`, not `expired`. You can pass identity_id instead of credential_id and the server picks that identity's credential.

• wire_login — sign in to a credentials-mode site and get a usable

credential_id immediately. Pass catalog_slug and params matching that catalog's login_input_schema (from wire_catalog). The password is never stored — only an encrypted session.

• If the user keeps the login in a 1Password identity source connected to

their Anakin account, pass source_id and source_ref ({ vault_id, item_id, fields }) instead of params, plus an identity_name, which is required in this mode. No password passes through the conversation.

Not every site supports password sign-in. Cookie-based sites use the dashboard connect flow instead; when that applies, the error includes a connect_url to give the user.

Never invent credentials. Ask the user, or send them to the connect_url.

When no action exists

wire_build requests a brand-new action: pass website_url and a specific goal describing what to extract or do. Wire generates and auto-tests a scraper, then publishes it.

• Async — returns pending.
• Charges credits, automatically refunded if the build fails.
• visibility defaults to private. Pass catalog_id to add the action to an

existing catalog instead of creating one.

• Rejected with ACTION_EXISTS if similar actions exist; pass force: true to

override.

Three more options:

• actions — a list of discrete capabilities, each built as its own action,

e.g. ["search products", "get product details"]. Omit it to let the builder infer them from goal.

• country — a two-letter code, e.g. "US", for a site that shows different

content by country. The action is built and tested through that country.

• credential — builds actions behind a sign-in. **A login build costs

significantly more credits than a public one**, so say so and get the user's go-ahead first. Shapes: { type: "plain", username, password }, or { type: "vault", source_id, source_ref } for a connected 1Password entry, either with an optional login_url. The password is used once and never stored. Never invent credentials; ask the user.

Track a build with wire_build_status:

• Pass the id from build_request and poll every ~30s while it is pending

or processing. The finished build lists each published action's action_id.

• Read the `skipped` list when it finishes. A build can deliver only part

of what was asked.

• Omit id to list recent builds, optionally filtered by status.
• include_events: true adds the step-by-step build log. It is verbose; use it

only to explain a failed build.

Only call this after `wire_discover` and `wire_catalog` confirm nothing covers the site. Building duplicates wastes credits and time.

Choosing Wire vs. the alternatives

Rule of thumb: if you could name the site to a colleague and they would recognize it, check Wire before scraping it. Wire actions are faster and cheaper than browser_task — always check wire_discover before driving a browser.