Loading ToolBench
Fiasco is a headless orchestrator with 5 tools. Tool definitions are inferred from test files rather than explicit registration code visible in the source. Schemas are present but minimal; descriptions are present but generic. The bash and mcp tools are particularly risky (IRREVERSIBLE) with minimal guidance on error recovery. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite high-risk operations. Parameter descriptions lack format constraints, ranges, or validation rules. Output schemas are not documented. Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors.
Execute bash commands in the workspace
Load and register a skill from the skill catalog
Execute MCP (Model Context Protocol) tool calls
Read file contents with line and byte offset support, including image attachment
Create or edit files with atomic text replacements
Tool definitions inferred from test files, not explicit registration visible in source.
bash and mcp tools marked IRREVERSIBLE with no error recovery guidance, confirmation steps, or dry-run support. Agents cannot self-correct on failure.
No tool annotations (destructiveHint, readOnlyHint, idempotentHint) despite write, bash, and mcp being destructive. LLMs cannot reason about side effects.
Parameter descriptions lack format constraints, ranges, or validation rules. E.g., 'command' has no length limit, timeout, or shell safety guidance. 'path' has no traversal boundary.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | <=2025-11-25 | v2 |
Output schemas not documented. Agents cannot plan downstream tool calls or extract required fields (e.g., what does bash return on success vs error?).