A Go library that lets an LLM agent orchestrate its own tools from a JavaScript program instead of one tool call per turn. The model submits a program as a single tool call where `await tools.name(args)` calls bound tools, `Promise.all` fans them out in parallel, and JavaScript merges/filters/scores results.
Single tool 'run_code' has a well-structured schema with required parameters (code, description) and clear types. Description is comprehensive (194 chars) and explains the tool's purpose, when to use it, and what it orchestrates. However, the description is overly long for LLM parsing and lacks actionable error guidance. Parameter descriptions are present but could be more prescriptive about constraints (e.g., code length limits, description format rules). No output schema is documented, the tool returns whatever the JavaScript program prints/returns, which is opaque to the LLM. Risk classification (READ_ONLY) is present but tool lacks idempotent/destructive hints. Error handling is minimal, no guidance on what happens when code fails, times out, or calls blocked tools.
Runs one JavaScript program that orchestrates your tools in batch; only what the program prints or returns comes back to the conversation. The tools callable in a program are exactly those in your current tool list, minus: run_code. Reach for it when the steps between calls are computable from the results themselves — parallel fan-out, chained transforms, branching or retrying mid-program, filtering bulk results down to what matters; a lone call is cheaper made directly.
No output schema documented. LLM cannot predict what the tool returns, making it impossible to plan downstream tool calls or extract structured data from results.
Parameter 'code' lacks constraints: no length limit, complexity limit, or timeout guidance. LLMs may generate arbitrarily large or slow programs.
Error handling is implicit. No guidance on what happens when code throws, times out, or calls a blocked tool. LLM cannot self-correct or retry intelligently.
Tool description is 194 chars, at the upper bound of the 10 - 1024 range. Verbose phrasing ('Reach for it when...') wastes tokens. Condense to ~100 chars for clarity.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2025-06-18+ | v2 |
Parameter 'description' lacks format guidance. Should specify: 5 - 10 words, active voice, no examples. Currently relies on prose explanation.