# Implement business-specific output files Build one registered feed with a separate filtered output for each business key. Use this skill when implementing target discovery, lookup, query filtering, and filenames. Produce a complete targeted feed, selection examples, and checks that target datasets and paths stay independent. ## Workflow 1. Identify the registration, business key, selection conditions, output contract, and expected target count. Choose targets for business variants, `perFile()` for count-based parts, or both when each variant needs parts. 2. Inspect the existing feed and callers. Keep one registration for the class. Implement `HasFeedTargets` with `InteractsWithFeedTargets`; keep `targets()` lazy and make `findTarget()` resolve explicit keys directly. 3. Design stable non-empty keys and small serializable parameter snapshots. Keys are visible in output and errors and are not sanitized. Use controlled path-safe keys, identifiers instead of models, and no secrets. 4. Use `target()->parameters` to filter `builder()` and a deterministic distinct `filename()` for every target. Return a lazy Eloquent builder. Resolve only required relationships and fields. 5. Exercise full enumeration, selected-key lookup, unknown keys, clone independence, dataset filtering, and filename uniqueness. For parts, verify suffixes and total counts independently per target. 6. Show the registration slug and repeatable `--target` commands. For direct service calls, resolve a target and pass `forTarget($target)`; `GeneratorService` never enumerates targets. Use [the target contract](references/TARGET-CONTRACT.md) for execution and failure rules. Use [the complete locale example](examples/LOCALE-FEED.md) for code, invocation, and a small assertion check. ## Artifact ownership and config policy Read current feed definitions, model schema, registrations, application storage/config, and call sites. Write only requested target implementation and relevant application tests or reproducible registration changes. Do not create one registration per target. Generation and queue dispatch change application state; perform them only within the requested scope. Retain the existing format, disk, and unrelated query rules unless the output contract requires a change. Do not expose secrets in keys, snapshots, logs, or filenames. This skill does not write AI Factory context or config.

Implement business-specific output files

Build one registered feed with a separate filtered output for each business key. Use this skill when implementing target discovery, lookup, query filtering, and filenames. Produce a complete targeted feed, selection examples, and checks that target datasets and paths stay independent.

Workflow

Identify the registration, business key, selection conditions, output contract, and expected target count. Choose targets for business variants, perFile() for count-based parts, or both when each variant needs parts.
Inspect the existing feed and callers. Keep one registration for the class. Implement HasFeedTargets with InteractsWithFeedTargets; keep targets() lazy and make findTarget() resolve explicit keys directly.
Design stable non-empty keys and small serializable parameter snapshots. Keys are visible in output and errors and are not sanitized. Use controlled path-safe keys, identifiers instead of models, and no secrets.
Use target()->parameters to filter builder() and a deterministic distinct filename() for every target. Return a lazy Eloquent builder. Resolve only required relationships and fields.
Exercise full enumeration, selected-key lookup, unknown keys, clone independence, dataset filtering, and filename uniqueness. For parts, verify suffixes and total counts independently per target.
Show the registration slug and repeatable --target commands. For direct service calls, resolve a target and pass forTarget($target); GeneratorService never enumerates targets.

Use the target contract for execution and failure rules. Use the complete locale example for code, invocation, and a small assertion check.

Artifact ownership and config policy

Read current feed definitions, model schema, registrations, application storage/config, and call sites. Write only requested target implementation and relevant application tests or reproducible registration changes. Do not create one registration per target. Generation and queue dispatch change application state; perform them only within the requested scope. Retain the existing format, disk, and unrelated query rules unless the output contract requires a change. Do not expose secrets in keys, snapshots, logs, or filenames. This skill does not write AI Factory context or config.