Apify Skill

Run any Apify Actor through a standardized workflow: search → validate → execute → collect results.

Prerequisites

• - APIFY_TOKEN env var, or a config.json with tokens (copy config.json.example)
• Python 3 with requests installed

Workflow

Step 1: Parse User Intent

Extract from the user's request:

• - Platform/target (Instagram, TikTok, Reddit, etc.)
• What to scrape (posts, profiles, hashtags, comments, etc.)
• Targets (URLs, usernames, keywords)
• Quantity/filters (how many, time range, min likes, etc.)

Step 2: Select Token

If user specifies a token name or the task maps to a specific account, use that. Otherwise use default.

Token can be provided via:

1. 1. --token flag (highest priority)
2. INLINECODE6 tokens map (by --token-name)
3. INLINECODE8 env var (fallback)

Step 3: Search & Select Actor

Run the search script:

CODEBLOCK0

Output: ranked candidates with score, success rate, rating, pricing model.

Quality filters (built into script):

• - notice = NONE (not deprecated)
• 30-day success rate ≥ 95%
• 30-day runs ≥ 1,000
• User rating ≥ 4.0

Pick the top-ranked candidate. If user has a preference or prior experience with a specific Actor, skip search.

Step 4: Get Actor Schema & Build run_input

Fetch the Actor's documentation:

CODEBLOCK1

Read the input schema section. Construct run_input JSON based on:

• - The Actor's required/optional fields
• The user's targets and filters
• Sensible defaults from the documentation

Do NOT ask the user to write JSON. Build it from their natural language request.

Step 5: Probe Test (Top 1 → Top 2 → Top 3 fallback)

Test with minimal input before committing to full run:

CODEBLOCK2

The probe automatically uses the first 2 items from the list field.

Checks:

• - Run starts successfully (no permission/billing errors)
• Run completes (no timeout/crash)
• Returns non-empty data

If probe fails → try next candidate Actor. If all 3 fail → report to user with Actor URLs for manual activation.

Step 6: Full Execution

CODEBLOCK3

Key flags:

Batching rules:

• - ≤ batch-size items → single run
• \> batch-size items → auto-split, 3s pause between batches
• Each batch has independent timeout (default 10 min)

Step 7: Return Results

• - Report total items collected
• Save raw JSON to specified output path
• Summarize key stats (items count, batches, any failures)
• Let the caller handle filtering/reporting/delivery

Common Actor Patterns

These are starting points. Always verify with the Actor's .md page for current schema.