AI 图像生成 by CreatOK

概述

generate-image

This skill is a thin client over the creatok CLI. The CLI owns all interface detail (auth, upload, submission, polling, artifacts, error normalization). The skill expresses intent, runs CLI commands, and reads their JSON output.

Constraints

• The model's final user-facing response should match the user's input language, default English.
• Must request user confirmation before triggering any paid/high-cost image generation call.
• After confirmed, must run the creatok image command and wait until completion.
• Preserve any explicit user-provided prompt verbatim. Follow Explicit Prompt Preservation in ./references/common-rules.md; generation confirmation never authorizes prompt changes.
• Avoid technical wording in the user-facing reply unless the user explicitly needs details for debugging.
• Follow shared guidance in ./references/common-rules.md.

The creatok CLI

All API interaction goes through the CLI. Never call HTTP endpoints directly.

• Auth is channel-specific: stable CLI uses CREATOK_API_KEY; preview CLI uses CREATOK_API_KEY_PREVIEW. These keys are parallel, and preview does not read or fall back to CREATOK_API_KEY.
• Every command prints a single JSON envelope to stdout; human logs go to stderr.
• Envelope shape: { "ok", "cli_version", "command", "task_id", "status", "data", "artifacts", "update", "error" }.
• On error, error.kind ∈ { auth, notfound, invalid, server, timeout, network, outdated_cli }.

Commands used by this skill:

creatok doctor                        # check key / reachability / skill drift / updates
creatok capabilities                  # models, defaults, hard limits
creatok assets create --type image --file "<path>"
creatok image --prompt "<p>" (--options '<json>' | --options-file options.json) --out <dir>
creatok status --task <id> --type image_generation [--wait]

Variable parameters (model, resolution, n, aspect_ratio, …) are passed through --options '<json>', e.g. --options '{"model":"nano-banana-2","resolution":"2K","n":2,"aspect_ratio":"16:9"}'. Use --options-file options.json when the shell makes inline JSON quoting unreliable, especially in PowerShell. Do not invent flags; anything beyond --prompt goes inside --options.

Model Selection Rules

Read model list, defaults, and hard limits from creatok capabilities at runtime instead of hardcoding them.

• Use the capabilities default_model unless the user explicitly chooses another supported model.
• Validate only hard execution constraints (resolution, aspect ratio, n, max reference images) against the model's limits.
• Keep user-facing model explanation minimal.

Inputs to clarify (ask if missing)

• ask only for what is necessary to generate a good image
• resolve model, resolution, image count (n), and aspect ratio from the user's choices or capabilities defaults, then pass all four explicitly
• if the prompt is vague, offer to refine it and wait for explicit permission; otherwise submit it unchanged
• reference images are optional — ask only if the user implies style transfer or subject reference
• for each local reference image, run creatok assets create --type image --file "<path>" and parse data.asset.object_key
• collect the returned formal Asset keys in imageObjectKeys; do not pass those same local files through --ref

Workflow

1. Entry self-check

• Run creatok doctor.
• If the command is not found, the CLI is not installed — guide the user to install it (see ./references/common-rules.md) and stop.
• If data.api_key_configured is false or error.kind is auth, guide the user to configure the channel-specific key: CREATOK_API_KEY for stable, CREATOK_API_KEY_PREVIEW for preview.
• Check for a newer CLI and, if the user agrees, update before proceeding — follow ./references/common-rules.md (Outdated CLI).
• If data.skills_drift is true, the installed skills are out of sync with the CLI — run creatok skills install to resync (see ./references/common-rules.md).

1. Resolve parameters

• Run creatok capabilities and pick the default image model + its limits.
• Build a strict --options JSON containing model, resolution, n, and aspect_ratio. If the selected model exposes pricing.quality_options, choose and pass one supported quality explicitly; otherwise omit quality. Add imageObjectKeys only when references are used.
• For each local reference image, run creatok assets create --type image --file "<path>".
• Parse each formal key from data.asset.object_key, preserve input order, and put the collected keys in imageObjectKeys.
• Asset creation happens before paid-task confirmation. Each created Asset remains in the user's media library and counts toward storage even if the user does not submit image generation.

1. Confirmation gate (mandatory, conversation layer)

• Summarize: model, resolution, number of images (n), aspect ratio, quality when the selected model supports it, and estimated cost/credits from capabilities pricing if available.
• Ask for a simple confirmation. Do not run creatok image until the user says yes.

1. Generate

• Run: creatok image --prompt "<prompt>" (--options '<json>' | --options-file options.json) --out <run_dir>.
• Use the exact approved prompt text for <prompt>. Do not optimize, rewrite, translate, expand, or correct it during command construction.
• When local reference images were created as Assets, submit their same formal keys together with all required fields, for example --options '{"model":"nano-banana-2","resolution":"2K","n":1,"aspect_ratio":"1:1","imageObjectKeys":["<asset_key_1>","<asset_key_2>"]}'.
• Do not also pass --ref for those files; that would create a second Asset for each local reference.
• The CLI submits, polls to completion, and persists artifacts.

1. Respond

• Parse the stdout envelope. On success, the final image URLs are at data.result.images[].url; treat them only as download inputs.
• Follow Mandatory Image-Only Result Response in ./references/common-rules.md: download and verify every successful image, render the required local Markdown image directives, and include the exact 打开图片所在位置 hyperlink to their containing directory.
• The CLI also writes result.json under artifacts.dir for recovery.
• If error is present, map error.kind to the guidance in ./references/common-rules.md and explain plainly.

Existing Task Recovery

• If the user already has a task_id, run creatok status --task <id> --type image_generation instead of starting a new generation.
• In recovery mode, do not ask the user to restate the prompt.
• Add --wait to keep polling if the user wants to wait for a still-running task.
• If succeeded, follow Mandatory Image-Only Result Response in ./references/common-rules.md; never return the remote image URLs directly.
• If still queued/running, explain clearly and offer to keep checking.
• If failed, explain the failure and suggest next steps.

Thin Client Boundary

• This skill only expresses intent and reads CLI JSON. It does not know endpoints, request bodies, or polling mechanics.
• The model should not make the user restate their idea if the direction is already clear from the conversation.