feat: add .actorignore to all templates and populate dataset_schema.json fields

[DaveHanns]

Summary

Library-wide hygiene + observability fix for templates created via apify create or fetched via @apify/actor-templates. Three coordinated changes:

1. .actorignore added to all 44 templates so apify push excludes build artifacts, dependencies, local storage, and dev/IDE files from the uploaded archive. Currently the entire library ships .dockerignore but zero templates ship .actorignore — every CLI-deploying agent or human pays for the gap.
2. .actor/dataset_schema.json added to 6 data-producing templates that materially push to a dataset but didn't ship a starter schema (ts-empty, js-empty, python-empty, ts-bootstrap-cheerio-crawler, js-bootstrap-cheerio-crawler, python-scrapy).
3. .fields populated in the 29 existing dataset_schema.json files (they were shipping "fields": {} empty alongside a populated views.overview.transformation.fields — the schemas were structurally incomplete). Derived .fields JSON Schema from each schema's existing field names + display-format hints.

Scope summary

Change	File count
.actorignore added	44
.actor/dataset_schema.json added	6
.actor/dataset_schema.json .fields populated	29
Total	79

Total diff: 79 files changed, 2311 insertions(+), 53 deletions(-).

.actorignore — language-appropriate contents

Three variants depending on the template's language:

• Node (TS / JS — 25 templates): dist/, build/, *.tsbuildinfo, node_modules/, storage/, apify_storage/, crawlee_storage/, *.log, .env*, IDE/OS files, .git/.
• Python (18 templates): __pycache__/, *.pyc, *.pyo, .venv//venv//env/, storage dirs, *.log, .env*, IDE/OS files, .git/.
• Minimal (cli-start, shell-only): *.log, .env*, IDE/OS files, .git/.

Templates intentionally NOT given a dataset_schema.json

Audited each template's main code file for Actor.pushData / Actor.push_data / equivalent. The following 9 templates do not push to a dataset and intentionally do not ship one:

• cli-start — shell-only template; no data path.
• js-langchain — writes only to the key-value store via Actor.setValue('OUTPUT', res). No pushData calls anywhere. (Already ships key_value_store_schema.json and output_schema.json — which is correct for its pattern.)
• ts-standby, js-standby, python-standby — HTTP servers; request/response, no dataset writes.
• ts-mcp-empty, ts-mcp-proxy, python-mcp-empty, python-mcp-proxy — MCP servers; serve endpoints, no dataset writes.

Shipping a dataset_schema for a non-dataset-writing template would imply a contract the template doesn't use and would mislead users about output location.

.fields population approach

For the 29 existing dataset_schema.json files with empty .fields, derived fields.properties from the schema's existing views.overview.transformation.fields list, with types mapped from the corresponding views.overview.display.properties.<name>.format:

Display format	Derived JSON Schema
link, image	{ "type": "string", "format": "uri" }
array	{ "type": "array", "items": { "type": "string" } }
number	{ "type": "number" }
boolean	{ "type": "boolean" }
date	{ "type": "string", "format": "date-time" }
object	{ "type": "object", "additionalProperties": true }
text / default	{ "type": "string" }

Each populated schema also gains a "required": [...] listing all derived field names. The structure remains internally consistent with what the existing views already declared.

What this fixes downstream

Closes the following items from the chocholous/apify-evals finding tracker:

• F8 / F27 — agents and humans ship bloated archives (no .actorignore).
• F29 — agents ship dist/ alongside src/ because the Dockerfile doesn't get a chance to rebuild from scratch.
• F26 — ts-empty (and other empty starters) ship no dataset_schema.json, triggering T1 soft-warns on every template-using agent.
• F3 — .fields empty soft-warn fires on every template-derived Actor.

Out of scope (separate PRs)

• The 16 JS/TS templates where name is fundamentally different from id (e.g. js-empty / project_empty, js-start / getting_started_node) warrant a coordinated rename + apify-cli alias support to maintain backward compatibility. Tracked separately.
• The @apify/json_schemas actor.schema.json .version regex being looser than the platform's MAJOR_MINOR_VERSION_REGEX is an apify-shared-js issue, not a templates one.

Test plan

• CI green on existing tests
• Spot-check apify create -t ts-empty produces a scaffold that includes both .actorignore and .actor/dataset_schema.json after dist/templates/ts-empty.zip is regenerated by CI
• Spot-check apify push archive size shrinks for a template scaffold (no node_modules/ or storage/)

🤖 Generated with Claude Code