Loading ToolBench
Command-line interface for Callirra API — generate images and video, browse 172 curated prompt recipes, and manage your account from the terminal
Callirra CLI exposes 15 tools via STDIO with basic naming and descriptions, but lacks critical production-grade patterns. Tool names are inconsistent (setup-api-key vs gen image vs cancel); descriptions are terse (10-50 chars, well below the 194-char baseline); parameter schemas are present but minimal (no enums, ranges, or validation guidance); no output schemas documented; no error recovery guidance; no tool annotations (readOnlyHint/destructiveHint). The server reads from local JSON files for recipes/scenes but does not expose these as resources. No pagination support despite tools returning lists. Security concern: setup-api-key accepts raw API keys as parameters (should use server-side injection). Overall, this is a thin CLI wrapper with minimal agent-optimization.
Show balance and available credits
Cancel a queued/running video task
Show creative knowledge / resource data
Generate an image
Generate a video
List available models
Show one recipe: prompt, settings, credits, deep link
Credentials exposed as tool parameters: setup-api-key accepts raw API keys as input, risking leakage into logs and traces. Should use server-side secret injection via environment variables.
Descriptions are too terse (10-50 chars vs 194-char baseline). LLMs cannot determine when to select tools or how to use them. E.g., 'Show one video task' lacks context on what data is returned or when to call it.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields. E.g., models, balance, usage, task, videos return data but structure is undocumented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
List the curated prompt library (172 recipes)
Show the six worked Seedance 2.5 scene recipes
Save an API key locally
Show one video task
Upload a reference image
Show recent usage (default 20)
List video tasks (alias: list-videos)
Show configured key and current balance
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe reads from destructive writes. E.g., cancel and gen video are reversible/destructive but unmarked.
No error recovery guidance. Tools fail silently or with generic errors. E.g., if an API key is invalid, the error should suggest 'Run setup-api-key with a valid key starting with sk-cal-'.