01Safety
0.0 / 30
What changed in the harness
Selection accuracy 100→100, token cost up 10%, unconfirmed writes 0%→0%.
Category breakdown
Where the score comes from.
Earned points across the four signals Gradable measures. Safety and Legibility are scored out of 30; Economics and Discoverability are scored out of 20.
02Legibility
26.3 / 30
03Economics
19.8 / 20
04Discoverability
9.9 / 20
Highest-impact fix
Estimated gain +30 pointsAdd explicit identity and permission preflight tools
Expose machine-readable principal/tenant confirmation and a non-mutating permission check so agents can verify both before destructive actions.
Description evidence
Defects and rewrites.
6 defects found across the exposed tool descriptions. Suggested rewrites make purpose, inputs, boundaries, and returns easier for an agent to understand.
| Tool | Defect types | Suggested rewrite |
|---|---|---|
list_models |
name_restates_behavior no_return_description |
Query the Pollinations catalog of supported generation models across image, video, and audio, including each model's type and pricing. Use it to select a valid model ID and understand its cost before generating; returns the full list of available models with their type and pricing information. |
list_styles |
name_restates_behavior no_return_description |
Retrieve the complete set of prompt-building vocabulary grouped by category, covering styles, lighting, camera angles, moods, colors, and quality tags. Use it to discover valid tag options for build_prompt; returns the categorized list of all available tags. |
build_prompt |
name_restates_behavior no_return_description |
Combine a required subject with optional style, lighting, camera, mood, color, and quality tag selections into a single optimized prompt string. Call this before a generation tool to ensure a well-formed prompt; returns the assembled prompt text ready for generate_image or generate_video. |
generate_video |
params_unexplained |
Generate a video clip via the Pollinations API and return a URL to the result. aspect_ratio sets the video frame ratio as width:height, choosing one of 1:1, 16:9, or 9:16; duration specifies length in seconds, which may be capped depending on the chosen model; seed makes output reproducible. |
check_balance |
name_restates_behavior no_return_description |
Inspect the Pollinations API credit balance (pollen) to see remaining credits for paid models. Query this before running paid generation workloads to avoid failures when credits run out; returns the current balance with remaining credit details. |
generate_batch |
no_return_description |
Generate up to 10 images in a single request, which is more efficient than calling generate_image repeatedly. Takes an array of up to 10 prompt objects, each with a required prompt and optional model, width, height, and seed; returns the generated image URLs for the batch. |
Selection evidence
Confusable tool pairs.
5 pairs where similar names or overlapping descriptions may send an agent toward the wrong tool.
| Tool A | Tool B | Confidence | Why they collide |
|---|---|---|---|
generate_image |
generate_batch |
medium | A task like 'generate 5 images of cats' fits both tools: generate_batch exists precisely to avoid calling generate_image repeatedly, but an agent may pick generate_image and loop, or pick generate_batch for a single-image intent. The neighbor-the-boundary use case makes the choice genuinely ambiguous. |
generate_image |
generate_video |
medium | Tasks that omit the medium ('make a clip of a whale', 'create an animation of a dancing robot', 'generate a scene of a sunset') don't clearly say image vs video, so an agent could pick either generate tool despite the descriptions distinguishing them. |
list_models |
list_styles |
medium | Both are zero-arg listing tools, so vague discovery tasks like 'what can you generate?', 'show me all available options', or 'what are my choices?' could plausibly resolve to either tool depending on whether the agent interprets 'options' as models or styles. |
list_styles |
build_prompt |
medium | Tasks about prompt construction with style options ('I need a prompt with cinematic lighting and mood', 'help me pick options for a prompt', 'give me a dramatic, colorful prompt') overlap heavily in vocabulary and could be answered by listing style options or by building the prompt itself. |
generate_image |
generate_audio |
low | Tasks like 'generate a tone' or 'generate a voice clip' are usually medium-specific, but words like 'clip', 'sound' misused, or a task asking to 'generate a bird' for a song vs a picture creates a small but real chance of confusion. |
Compare the field