Two workflows that chain primitives, composed from the skills' own commands

price-monitoring (fetch -> extract -> batch): prove the page, prove the
structure, estimate before spending, run managed, retry-failed on a
schedule. gated-page-extraction (fetch -> browser -> extract): the
escalation-only rule end to end - the browser enters only when the cheap
step failed with evidence, scripted sessions preferred because they
close themselves.

Every command line is lifted from the skills (protected-fetch, extract,
batch-jobs, interact-browser), so the workflows cannot promise a flag
the CLI does not have. Registered in registry/workflows.json; the app's
Overview vendors this registry (scripts/sync-agent-catalog.mjs there),
so these cards render in the product from this same source.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: pablomogamon

registry/workflows.json

@@ -5,11 +5,58 @@
       "type": "workflow",
       "description": "Gather competitor pages, extract structured data, and keep it fresh.",
       "status": "available",
-      "requires_backend_capabilities": ["protected_fetch", "extract"],
+      "requires_backend_capabilities": [
+        "protected_fetch",
+        "extract"
+      ],
       "requires_auth": true,
       "version": "0.1.0",
       "path": "workflows/competitor-intelligence",
-      "tags": ["competitive", "extract", "fetch"]
+      "tags": [
+        "competitive",
+        "extract",
+        "fetch"
+      ]
+    },
+    {
+      "name": "price-monitoring",
+      "type": "workflow",
+      "description": "Watch prices across many product pages, validated once and then run as a managed batch.",
+      "status": "available",
+      "requires_backend_capabilities": [
+        "protected_fetch",
+        "extract",
+        "batch"
+      ],
+      "requires_auth": true,
+      "version": "0.1.0",
+      "path": "workflows/price-monitoring",
+      "tags": [
+        "pricing",
+        "fetch",
+        "extract",
+        "batch"
+      ]
+    },
+    {
+      "name": "gated-page-extraction",
+      "type": "workflow",
+      "description": "Get structured data out of pages that need interaction first, escalating to a browser only with evidence.",
+      "status": "available",
+      "requires_backend_capabilities": [
+        "protected_fetch",
+        "browser",
+        "extract"
+      ],
+      "requires_auth": true,
+      "version": "0.1.0",
+      "path": "workflows/gated-page-extraction",
+      "tags": [
+        "browser",
+        "escalation",
+        "extract",
+        "fetch"
+      ]
     }
   ]
 }

workflows/gated-page-extraction/WORKFLOW.md

@@ -0,0 +1,51 @@
+---
+name: gated-page-extraction
+description: Get structured data out of pages that need interaction first, escalating to a browser only with evidence.
+requires_backend_capabilities: [protected_fetch, browser, extract]
+status: available
+---
+
+# Workflow: gated-page extraction
+
+Some pages only show their data after an interaction: a login, a "load more",
+a form. The rule from [[interact-browser]] applies end to end: **escalation
+only** — a browser session costs more than fetch/extract (billed by bandwidth
+plus session time, 15-minute cap), so it enters the flow only when the cheap
+step has failed with evidence.
+
+## Steps
+
+1. **Try the cheap primitive first** ([[protected-fetch]]):
+   ```
+   zenrows fetch https://portal.example/reports
+   ```
+   If the response already carries the data, stop here — the rest of this
+   workflow is unnecessary cost. If it fails, read the trace before escalating
+   ([[trace-debug]]).
+2. **Escalate to a scripted session.** Put the interaction (navigate, type,
+   click, read) in a script and let the session close itself
+   ([[interact-browser]]):
+   ```
+   zenrows browser run script.json
+   ```
+   Prefer the scripted form over an interactive session: interactive sessions
+   you must `close` yourself, and every open minute bills.
+3. **Turn the captured page into data** ([[extract]]):
+   ```
+   zenrows extract https://portal.example/reports --autoparse
+   ```
+   Or a selector map when you know the shape:
+   ```
+   zenrows extract https://portal.example/reports --css '{"report":"h2","total":".sum"}' --validate
+   ```
+
+## Explain
+
+```
+zenrows workflow explain gated-page-extraction
+```
+
+Each step declares its capability; the toolkit refuses honestly when a required
+primitive is not available yet (`ASSET_REQUIRES_CAPABILITY`). Browser Sessions
+is GA and on by default; a workspace that opted out re-enables it with
+`zenrows policy set allow_browser true` ([[compliance-policy]]).

workflows/price-monitoring/WORKFLOW.md

@@ -0,0 +1,52 @@
+---
+name: price-monitoring
+description: Watch prices across many product pages, validated once and then run as a managed batch.
+requires_backend_capabilities: [protected_fetch, extract, batch]
+status: available
+---
+
+# Workflow: price monitoring
+
+Three primitives in a row: prove the page with **Fetch**, prove the structure
+with **Extract**, then let **Batch** operate the loop. Each step is the cheapest
+tool that answers the next question, per [[cost-control]].
+
+## Steps
+
+1. **Prove the page loads with its price.** One protected fetch, waiting for
+   the selector you care about ([[protected-fetch]]):
+   ```
+   zenrows fetch https://shop.example/p/123 --wait-for ".price"
+   ```
+2. **Prove the structure once.** A selector map that must come back as valid
+   JSON, or fail loudly ([[extract]]):
+   ```
+   zenrows extract https://shop.example/p/123 --css '{"title":"h1","price":".price"}' --validate
+   ```
+3. **Estimate before spending.** Put the full URL list in `jobs.jsonl` and ask
+   what the run would cost — this validates the spec locally and needs no key
+   ([[batch-jobs]]):
+   ```
+   zenrows batch estimate jobs.jsonl
+   ```
+4. **Run it managed.** Batch submits every URL, retries transient failures and
+   stores results, so you do not operate the loop yourself:
+   ```
+   zenrows batch create jobs.jsonl --wait
+   zenrows batch results <id> --status all --out results.jsonl
+   ```
+5. **Keep it fresh.** Rerun on a schedule (cron or CI). Failures do not restart
+   the world:
+   ```
+   zenrows batch retry-failed <id>
+   ```
+
+## Explain
+
+```
+zenrows workflow explain price-monitoring
+```
+
+Each step declares its capability; the toolkit refuses honestly when a required
+primitive is not available yet (`ASSET_REQUIRES_CAPABILITY`). Batch is beta —
+`estimate` runs locally today, the cloud steps need beta access.