docs(skills): merge output-schema skill into actor-development

[patrikbraborec]

Summary

Reviewed the Actor development skills against the writing-for-agents guidance and consolidated them so there is one skill to install and one place for schema rules.

Two skills become one, and four overlapping schema references collapse into one:

 skills/
 ├── apify-actor-development/
 │   ├── SKILL.md                          # 2,221 → 1,292 words, rewritten
 │   └── references/
 │       ├── input-schema.md               # 191 → 550 words, rewritten from v1 spec
-│       ├── dataset-schema.md             # −209
-│       ├── output-schema.md              # −49
-│       ├── key-value-store-schema.md     # −129
+│       └── output-schemas.md             # +152, the single source of schema rules
-├── apify-generate-output-schema/
-│   └── SKILL.md                          # −415, whole skill removed
 └── apify-actorization/
     └── references/schemas-and-output.md  # repointed at output-schemas.md

Why

The two skills disagreed, so an agent following one produced schemas the other rejected:

 agent writes a schema
-  reads actor-development/references/dataset-schema.md
-    → omits `nullable`, puts `required` at one level only
-  generate-output-schema skill validates it
-    → rejects: hard rules say `nullable` everywhere, `required` at both levels
+  reads actor-development/references/output-schemas.md
+    → one set of rules, no contradiction

The same disagreement covered additionalProperties at both levels and type: string on output properties.

What changed

• Merge apify-generate-output-schema into apify-actor-development as references/output-schemas.md. The standalone skill, its marketplace entry, and its README/AGENTS.md references are removed. Its trigger branch ("generate or update schemas") now lives in the actor-development description.
• Delete the three duplicate schema references (dataset-schema.md, output-schema.md, key-value-store-schema.md).
• Rewrite input-schema.md from the v1 specification: default vs prefill vs required, editors per type, isSecret, errorMessage, resource fields, deprecated patternKey/patternValue.
• Rewrite SKILL.md: workflow steps first with checkable done-conditions, reference pointers folded into the steps, each rule stated once in positive form, command list reduced to the non-obvious entries.
• Point the actorization schema reference at the shared output-schemas.md. agents/AGENTS.md regenerated with scripts/generate_agents.py.

Aligned with the templates and apify/actor-templates#929

Reading the skill next to the new template AGENTS.md as an agent would turned up places where following either one produced wrong code:

• Python standby crashed: the example used Actor.config.container_port, which the Python SDK does not have. It is Actor.configuration.web_server_port, as in python-standby. Local testing now goes through apify run (checked: it serves js-standby on 4321 and answers the readiness probe with 200) instead of contradicting the "only apify run" rule.
• Logger: log from the apify package (import { Actor, log } from 'apify'), as every template imports it, instead of apify/log.
• Python layout: entry point my_actor/main.py, not src/main.py; uv projects (pyproject.toml, uv.lock) covered.
• actor.json keys: inputSchema / outputSchema, which the templates use since actor-templates#822; input / output are the deprecated names.
• Validation: apify validate-schema is part of the output-schema checklist. CLI 1.10 reads only the deprecated output key, so it skips an outputSchema; the checklist says so. A template's "fields": {} is a placeholder, not a style to match.
• Added: aborting handler (JS and Python), proxy setup from the proxyConfiguration input, a monetization section (pay-per-event Actor.charge, spending limit), apify actor generate-schema-types for TypeScript, Python crawler and router names.
• commands/create-actor.md reduced to discovery and approval gates around the skill's workflow. It linked to template files that do not exist and carried its own divergent workflow and README list.

Everything else in the diff is fallout from the removal:

graph LR
    A["delete apify-generate-output-schema"] --> B["marketplace.json<br/>−16 entry"]
    A --> C["README.md<br/>−4 listing"]
    A --> D["commands/create-actor.md<br/>repoint links"]
    A --> E["agents/AGENTS.md<br/>regenerated by script"]

Loading

File	Words before	Words after
apify-actor-development/SKILL.md	2,221	1,292
Output-schema material (skill + 3 references)	3,406	988
input-schema.md	191	550

Breaking change

The apify-generate-output-schema plugin no longer exists. Users who installed it get the same behaviour from apify-actor-development.

Dropped and left alone, flagging for review

dropped (could not verify against current Crawlee — restore if still relevant)
├── requestHandlerTimeoutMillis gotcha
└── additionalHttpHeaders vs preNavigationHooks

dropped (restates the environment and public docs)
├── project-structure tree
└── Playwright MCP JSON config block

left alone (pre-existing, not introduced here)
└── actor-json.md                   → still duplicates Structure/Example blocks and the `generatedBy` instruction