Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Base-layer skill for the SenseNova-Skills project, providing low-level APIs for image generation, recognition (VLM), and text optimization (LLM). This skill does not preprocess inputs; it only calls backend services and returns results. This skill is not user-facing and is intended for upper-layer skills only.
.claude/skills/opensensenova-sn-image-base/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 18 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 1336% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 136% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 922% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 346% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 130% | 0% |
bashpip install -r requirements.txt
sn-image-base is the base-layer skill (tier 0) of the SenseNova-Skills project and provides four low-level tools:
sn-image-generate: image generation (calls text-to-image-no-enhance API)sn-image-edit: image editing with SenseNova U1.5 Lite (calls /images/edits)sn-image-recognize: image recognition (uses VLM to analyze image content)sn-text-optimize: text optimization (uses LLM to process text)This skill does not perform any input preprocessing and only calls backend services to return results.
Image generation tool that calls the text-to-image-no-enhance API.
--prompt is required; all other parameters are optional:
| Parameter | Type | Default | Description | |------|------|--------|------| | --prompt | string | Required | Prompt text for image generation | | --negative-prompt | string | "" | Negative prompt | | --image-size | string | 2k | Image size preset (case-insensitive). Recommended: 2k. 4k is supported by sensenova-u1.5-lite; other SenseNova image models may reject it. Other values → status=failed. | | --aspect-ratio | string | 16:9 | Aspect ratio, e.g. 1:1, 16:9, 9:16 | | --seed | int | None | Random seed for reproducible generation | | --unet-name | string | None | Specify a UNet model name | | --api-key | string | SN_IMAGE_GEN_API_KEY -> SN_API_KEY | API key (CLI argument has priority; MissingApiKeyError is raised when all are empty) | | --base-url | string | SN_IMAGE_GEN_BASE_URL -> SN_BASE_URL | API base URL (CLI argument has priority) | | --poll-interval | float | 5.0 | Polling interval (seconds) | | --timeout | float | 300.0 | Timeout (seconds) | | --insecure | flag | False | Disable TLS verification | | --save-path | Path | Auto-generated | Save path |
SenseNova image requests explicitly send watermark=false by default. Both sensenova-u1-fast and sensenova-u1.5-lite are supported; U1.5 Lite additionally supports native 4K output. This no-watermark feature is currently in free public beta and may become paid.
Edits one or more reference images with SenseNova U1.5 Lite through the /images/edits endpoint. Local paths are converted to Data URLs; HTTP(S) URLs and Data URLs are passed through.
bashpython scripts/sn_agent_runner.py sn-image-edit \ --prompt "Change the background to a snowy mountain" \ --images source.png reference.png \ --save-path edited.png
The edit request uses the official defaults n=1, size=auto, watermark=false, prompt_extend=true, and response_format=url.
Image recognition tool that uses VLM (Vision Language Model) to analyze image content. Supports multiple image inputs.
--images and --user-prompt (or --user-prompt-path) are required. All other parameters use three-level defaults (CLI > env var > built-in default):
| Parameter | Type | Built-in Default | Env Var | Description | |------|------|-----------|---------|------| | --api-key | string | No hardcoded default | SN_VISION_API_KEY -> SN_CHAT_API_KEY -> SN_API_KEY | Chat runtime API key; raises MissingApiKeyError when all are unset | | --base-url | string | SN_CHAT_BASE_URL default | SN_VISION_BASE_URL -> SN_CHAT_BASE_URL -> SN_BASE_URL | Vision provider base URL; falls back to shared chat/global provider | | --model | string | sensenova-6.8-flash-lite | SN_VISION_MODEL -> SN_CHAT_MODEL | Vision-capable model name | | --vlm-type | string | openai-completions | SN_VISION_TYPE -> SN_CHAT_TYPE | Chat protocol type override | | --user-prompt-path | string | None | - | Local file path, mutually exclusive with --user-prompt | | --system-prompt-path | string | None | - | Local file path, mutually exclusive with --system-prompt |
Available values for --vlm-type:
openai-completions: OpenAI-compatible /v1/chat/completions interfaceanthropic-messages: Anthropic Messages /v1/messages interfaceText optimization tool that uses LLM (Language Model) to optimize text content. Does not accept image inputs.
--user-prompt (or --user-prompt-path) is required. All other parameters use three-level defaults (CLI > env var > built-in default):
| Parameter | Type | Built-in Default | Env Var | Description | |------|------|-----------|---------|------| | --api-key | string | No hardcoded default | SN_TEXT_API_KEY -> SN_CHAT_API_KEY -> SN_API_KEY | Chat runtime API key; raises MissingApiKeyError when all are unset | | --base-url | string | SN_CHAT_BASE_URL default | SN_TEXT_BASE_URL -> SN_CHAT_BASE_URL -> SN_BASE_URL | Text provider base URL; falls back to shared chat/global provider | | --model | string | sensenova-6.8-flash-lite | SN_TEXT_MODEL -> SN_CHAT_MODEL | Text model name | | --llm-type | string | openai-completions | SN_TEXT_TYPE -> SN_CHAT_TYPE | Chat protocol type override | | --user-prompt-path | string | None | - | Local file path, mutually exclusive with --user-prompt | | --system-prompt-path | string | None | - | Local file path, mutually exclusive with --system-prompt |
Available values for --llm-type:
openai-completions: OpenAI-compatible /v1/chat/completions interfaceanthropic-messages: Anthropic Messages /v1/messages interface| Tool | Model Type | Image Input | Interface Type Parameter | |------|----------|-----------------|-------------| | sn-image-recognize | VLM (Vision Language Model) | Yes, supports multiple images | --vlm-type | | sn-text-optimize | LLM (Language Model) | No, text only | --llm-type |
All tools are called through the unified sn_agent_runner.py entrypoint:
bash# Image generation (only prompt required; api-key/base-url have defaults) python scripts/sn_agent_runner.py sn-image-generate \ --prompt "..." # Image generation (override base-url) python scripts/sn_agent_runner.py sn-image-generate \ --prompt "..." \ --base-url "https://custom-endpoint.com/v1" # Image generation (explicitly override api-key) python scripts/sn_agent_runner.py sn-image-generate \ --prompt "..." \ --api-key "sk-xxx" # Image recognition (VLM) - minimal call (uses built-in Sensenova defaults) python scripts/sn_agent_runner.py sn-image-recognize \ --user-prompt "Describe the image" \ --images "path/to/image.png" # Image recognition (VLM) - override to Anthropic Claude API compatible (messages interface) python scripts/sn_agent_runner.py sn-image-recognize \ --user-prompt "Describe the image" \ --images "path/to/image.png" \ --api-key "sk-ant-xxx" \ --base-url "https://api.anthropic.com" \ --model "claude-sonnet-4-6" \ --vlm-type "anthropic-messages" # Text optimization (LLM) - minimal call (uses built-in Sensenova defaults) python scripts/sn_agent_runner.py sn-text-optimize \ --user-prompt "Optimize the text: ..." # Text optimization (LLM) - override to Anthropic Claude API compatible (messages interface) python scripts/sn_agent_runner.py sn-text-optimize \ --user-prompt "Optimize the text: ..." \ --api-key "sk-ant-xxx" \ --base-url "https://api.anthropic.com" \ --model "claude-sonnet-4-6" \ --llm-type "anthropic-messages"
Authentication parameters for sn-image-generate have the following default behavior:
| Parameter | Default | Override | Description | |------|--------|----------|------| | --base-url | SN_IMAGE_GEN_BASE_URL -> SN_BASE_URL | --base-url "..." | CLI argument has priority | | --api-key | SN_IMAGE_GEN_API_KEY -> SN_API_KEY | --api-key "..." | CLI argument has priority; throws MissingApiKeyError if all values are empty |
sn-image-recognize and sn-text-optimize use priority: CLI argument > command-specific env var > shared SN_CHAT_* env var > global SN_* env var > built-in default.
| Parameter | Built-in Default | Vision Env Var | Text Env Var | |------|-----------|-------------|-------------| | --api-key | None (must be provided) | SN_VISION_API_KEY -> SN_CHAT_API_KEY -> SN_API_KEY | SN_TEXT_API_KEY -> SN_CHAT_API_KEY -> SN_API_KEY | | --base-url | https://token.sensenova.cn/v1 | SN_VISION_BASE_URL -> SN_CHAT_BASE_URL -> SN_BASE_URL | SN_TEXT_BASE_URL -> SN_CHAT_BASE_URL -> SN_BASE_URL | | --model | sensenova-6.8-flash-lite | SN_VISION_MODEL -> SN_CHAT_MODEL | SN_TEXT_MODEL -> SN_CHAT_MODEL | | --vlm-type / --llm-type | openai-completions | SN_VISION_TYPE -> SN_CHAT_TYPE | SN_TEXT_TYPE -> SN_CHAT_TYPE |
api_key resolution order (high to low): CLI --api-key > command-specific key (SN_VISION_API_KEY/SN_TEXT_API_KEY) > SN_CHAT_API_KEY > SN_API_KEY. If all are unset, MissingApiKeyError is raised.
Only --api-key must be provided via CLI or environment; base URL, model, and interface type have shared chat defaults.
The agent can automatically read parameters from openclaw.json without manual input:
| CLI Parameter | openclaw.json Field | Example | |-----------|-------------------|--------| | --base-url | providers.<name>.baseUrl | https://api.anthropic.com | | --llm-type | providers.<name>.api | anthropic-messages / openai-completions | | --vlm-type | providers.<name>.api | anthropic-messages / openai-completions | | --model | providers.<name>.models[].id | claude-sonnet-4-6 | | --api-key | providers.<name>.apiKey or env var | sk-cp-... |
Note: --llm-type and --vlm-type share the same providers.<name>.api field and are used by LLM and VLM tools respectively.
Mapping between provider.api and interface type:
| api Value | Corresponding --llm-type / --vlm-type | Endpoint Path | |--------|----------------------------------|---------------| | anthropic-messages | anthropic-messages | /v1/messages | | openai-completions | openai-completions | /v1/chat/completions | | openai-responses | (future extension) | /responses |
Different API types have different requirements for base-url format:
| Type | --llm-type / --vlm-type | Recommended base-url | Code Appended Path | Final URL Example | |------|------------------------------|---------------|--------------|---------------| | LLM | openai-completions | https://token.sensenova.cn/v1 | /chat/completions | https://token.sensenova.cn/v1/chat/completions | | LLM | anthropic-messages | https://api.anthropic.com/v1 | /messages | https://api.anthropic.com/v1/messages | | VLM | openai-completions | https://token.sensenova.cn/v1 | /chat/completions | https://token.sensenova.cn/v1/chat/completions | | VLM | anthropic-messages | https://api.anthropic.com/v1 | /messages | https://api.anthropic.com/v1/messages |
Note:
/v1./v1/chat/completions or /v1/messages./v1, the runner appends only /chat/completions or /messages./v1, such as Gemini's /v1beta/openai.All tools support two output formats:
--output-format text (default): outputs plain text result--output-format json: outputs JSON, including status and elapsed_seconds (runtime in seconds, rounded to 2 decimals)JSON output for sn-image-recognize and sn-text-optimize also includes model, base_url, and interface_type to verify the effective runtime configuration:
json{ "status": "ok", "result": "...", "model": "sensenova-6.8-flash-lite", "base_url": "https://token.sensenova.cn/v1", "interface_type": "openai-completions", "elapsed_seconds": 1.23 }
On failure:
json{ "status": "failed", "error": "error message", "elapsed_seconds": 0.05 }
See references/api_spec.md for details.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,185 | 9,473 | -15% | 1 | 1 | 0% | 960 | 4,362 | +354% | 0 | 0 | — |
case-02 | fail→fail | 15,916 | 26,149 | +64% | 1 | 1 | 0% | 357 | 4,974 | +1293% | 0 | 0 | — |
case-03 | fail→fail | 9,029 | 43,281 | +379% | 1 | 1 | 0% | 718 | 9,178 | +1178% | 0 | 0 | — |
case-04 | pass→pass | 15,292 | 13,248 | -13% | 1 | 1 | 0% | 1,934 | 5,559 | +187% | 0 | 0 | — |
case-05 | pass→pass | 26,596 | 22,018 | -17% | 1 | 1 | 0% | 4,440 | 7,265 | +64% | 0 | 0 | — |
case-06 | pass→pass | 16,636 | 12,005 | -28% | 1 | 1 | 0% | 2,282 | 5,228 | +129% | 0 | 0 | — |
case-07 | fail→fail | 10,089 | 16,209 | +61% | 1 | 1 | 0% | 821 | 4,052 | +394% | 0 | 0 | — |
case-08 | pass→fail | 8,498 | 17,185 | +102% | 1 | 1 | 0% | 565 | 4,154 | +635% | 0 | 0 | — |
case-09 | fail→pass | 7,104 | 23,908 | +237% | 1 | 1 | 0% | 377 | 5,412 | +1336% | 0 | 0 | — |
case-10 | fail→pass | 23,986 | 6,952 | -71% | 1 | 1 | 0% | 1,776 | 4,199 | +136% | 0 | 0 | — |
case-11 | fail→pass | 8,046 | 10,140 | +26% | 1 | 1 | 0% | 476 | 4,864 | +922% | 0 | 0 | — |
case-12 | fail→pass | 13,681 | 26,013 | +90% | 1 | 1 | 0% | 1,434 | 6,401 | +346% | 0 | 0 | — |
case-13 | fail→pass | 17,172 | 9,632 | -44% | 1 | 1 | 0% | 2,045 | 4,701 | +130% | 0 | 0 | — |
case-14 | fail→pass | 16,213 | 7,989 | -51% | 1 | 1 | 0% | 2,127 | 4,329 | +104% | 0 | 0 | — |
case-15 | fail→pass | 19,707 | 13,857 | -30% | 1 | 1 | 0% | 2,482 | 5,733 | +131% | 0 | 0 | — |
case-16 | fail→pass | 23,064 | 8,193 | -64% | 1 | 1 | 0% | 3,318 | 4,487 | +35% | 0 | 0 | — |
case-17 | pass→pass | 16,483 | 7,650 | -54% | 1 | 1 | 0% | 2,178 | 4,355 | +100% | 0 | 0 | — |
case-18 | pass→pass | 17,518 | 7,500 | -57% | 1 | 1 | 0% | 2,009 | 4,277 | +113% | 0 | 0 | — |
case-19 | fail→pass | 16,926 | 7,980 | -53% | 1 | 1 | 0% | 1,931 | 4,330 | +124% | 0 | 0 | — |
case-20 | fail→pass | 41,787 | 8,639 | -79% | 1 | 1 | 0% | 1,948 | 4,358 | +124% | 0 | 0 | — |
case-21 | fail→pass | 16,392 | 6,856 | -58% | 1 | 1 | 0% | 2,218 | 4,121 | +86% | 0 | 0 | — |
case-22 | fail→pass | 19,310 | 7,632 | -60% | 1 | 1 | 0% | 2,377 | 4,304 | +81% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +50 percentage points is the difference between those two pass rates over the 19 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/24/2026 | +74% |
| gemini-3.6-flash | verified | 8/9/2026 | +68% |
Other measured skills in the registry, with their headline benchmark lift.