Two Actors, one agent, and the three ways my chain broke

Description and use

Two Actors, one agent, and the three ways my chain broke

Problem

A client needed to identify risky sellers on marketplaces, requiring two distinct steps: extracting sellers from listings and checking their legal entities against public risk registries. The existing two separate Actors were chained via an agent, but the chain failed in three subtle ways, producing incorrect results.

Outcome

The chain now correctly filters sellers without tax IDs, avoids permission errors, and produces accurate risk assessments.

Workflow

• 1. Configure the seller scraper Actor to output records with a 'checkable' flag and a 'checkableReason' field.
• 2. Update the input schema of the seller scraper to expose an 'items' array as the primary input, and demote 'sourceDatasetId' to a fallback.
• 3. In the risk checker Actor, add logic to skip records with empty tax IDs and push a 'skipped_no_inn' status.
• 4. Chain the two Actors via an agent, ensuring the agent passes the 'items' array by value.
• 5. Run the chain and verify that only sellers with valid tax IDs are processed and risk scores are returned.
• 6. Monitor logs for permission errors and adjust dataset access if necessary.

Evidence

The agent never sees the log. It saw a list of sellers, mapped it to the risk checker's input, and passed the whole thing. Records without a tax ID went in as empty strings.

Source

The fix is to pass values, and the part specific to chaining is this: your input schema decides which wiring the agent tries first.

Source

Second, the risk Actor refuses to invent an answer for input it cannot use.

Source