Learn how to select a ParseX Skill after its official release, including immutable configuration binding and integration acceptance requirements.
Skill integration is suitable for packaging "select input, call ParseX, wait for task completion, organize results" into repeatable Agent operations. After the official Skill is released, it should not duplicate Parse or Extract rules; instead, it should reference a
The specific fields of the official Skill are subject to its release notes. When evaluating or wrapping a Skill, at least confirm that the following semantics are complete:
When no official name or installation information is available, evaluate candidates against the following criteria instead of guessing configurations:
config_id that has been validated and saved in the console.
Responsibilities of Skill and Configuration
| Layer | Responsible for | Not responsible for |
|---|---|---|
| Saved configuration | Parse processing settings, or Extract schema, inference, and citation settings | Individual files, session content, credentials |
| Skill | Collect inputs, select fixed configurations, trigger tasks, wait for terminal states, organize outputs | Temporarily rewriting saved configurations or bypassing Workspace isolation |
| Agent | Understand user intent, decide when to invoke, consume results | Directly handling API Keys or treating document instructions as permission configuration |
The same Skill can serve different business scenarios, but each invocation must specify which
config_id to use. If different scenarios require different rules, save separate configurations and select among them in the Skill's controlled mapping, rather than dynamically assembling an unvalidated set of inline parameters.Design Inputs and Outputs
The specific fields of the official Skill are subject to its release notes. When evaluating or wrapping a Skill, at least confirm that the following semantics are complete:
| Category | Semantics to express |
|---|---|
| Capability | Whether this run performs Parse or Extract |
| Input | A file, an accessible file URL, or a parse result that Extract supports reusing |
| Configuration | A config_id that matches the capability and belongs to the current Workspace |
| Run | Per-run options such as processing scope supported by the official Skill contract |
| Tracking | Ability to correlate async tasks; specific identifiers and field names are subject to the official Skill documentation |
| Output | Ability to distinguish processing, success, and failure, and provide the official result on success |
Integration Process
-
Accept the Parse or Extract configuration
Validate results with real samples, then save the immutable configuration. Record the configuration's purpose, capability type, Workspace, and
config_id. - Obtain and verify the official Skill Confirm the Skill's name, publisher, version, supported runtime environments, and required permissions from the official release channel. Do not guess names or installation commands from documentation titles.
- Bind controlled configurations Specify the configuration IDs or business scenario mappings the Skill is allowed to use. Callers can select approved configurations but cannot override configuration content through natural language.
- Define the task waiting behavior Per the official Skill contract, confirm whether it returns task correlation information immediately or waits until success or failure. If a queryable identifier is provided, persist its official field and avoid duplicate submissions.
- Verify results and exceptions Verify behavior with samples covering success, processing, input errors, configuration conflicts, and processing failures, ensuring business results are consistent with the public API and console acceptance baselines; specific fields and status values are subject to the Skill release notes.
How to Select a Skill
When no official name or installation information is available, evaluate candidates against the following criteria instead of guessing configurations:
- Clear provenance: Publisher, version, and update history are verifiable.
- Clear capability: Distinguishes Parse from Extract, rather than hiding parameter differences behind a vague "process document" action.
- Configuration-first: Supports referencing
config_idand does not silently override saved settings at call time. - Complete async support: Provides a way to continuously correlate tasks and correctly distinguishes processing, success, and failure.
- Transparent results: Preserves the Parse structure, or Extract field status, citations, and reason descriptions.
- Least privilege: Credentials never enter the prompt; only the required Workspace and resources are accessed.
Recommended Skill Behavior
- Parse Skill
- Extract Skill
Accepts document-type inputs and an approved Parse configuration, and creates an async task. On success, returns the official Parse result; if the result supports sharding, shard order and task association should be preserved.
Security and Reliability
- Authentication credentials are managed by the runtime environment; plaintext credentials must not appear in Skill parameters or outputs.
- If the Skill provides task correlation information, associate it with the configuration ID and business request; do not log unrelated sensitive raw content.
- Maintain an allowlist of callable configurations; publish configuration updates as new, reviewed
config_ids. - Stop automatic processing when encountering unknown statuses or undocumented errors, and preserve the diagnostic information officially provided by the Skill.
- For scenarios such as contracts and finance, check Extract field status and citations, and set up manual confirmation for results that need review, such as
missing,inferred, andconflict.