Loading ToolBench
Self-hosted AI dev workflow control plane with native HTTP agent support, planning/task management, and API plugin orchestration
Kronn exposes 6 tools with basic schemas and descriptions. All tools have non-empty descriptions (10 - 200 chars, within baseline range) and input schemas with type definitions. However, descriptions are minimal and lack actionable guidance for LLM selection. Parameter descriptions are sparse or missing entirely. No output schemas are documented. Error handling is absent, tools do not guide recovery or categorize failures. Tool naming follows verb_noun convention (mcp_list, api_endpoints, qa_run, qe_run) but lacks clarity on when to use similar tools (qa_list vs qe_list distinction is unclear from names alone). No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles (qa_run and qe_run are WRITE operations). Composition is reasonable, tools chain logically (list → run), but discovery tools lack guidance on prerequisites.
List the endpoint paths one plugin exposes. Use the slug from mcp_list. Returns the paths you can pass to api_call.
List the API plugins available in this Kronn instance (slug + what each is for). Call this first to find a slug, then api_endpoints(slug) to see its paths.
List saved Quick APIs (pre-configured calls). Prefer running one of these over hand-building an api_call when it matches.
Execute a saved Quick API by id and return its result.
List saved Quick Execs — the commands this instance can run for you. Pass one's id to agent_job_start; there is no other way to learn it.
Run a saved Quick Exec now and return its output. Shell-free: the saved argv runs directly, bounded to its project. For a long one, prefer agent_job_start, which survives this turn.
No output schemas documented. LLMs cannot predict response structure, forcing them to guess at field names and types for downstream tool calls.
No tool annotations (readOnlyHint/destructiveHint). qa_run and qe_run are marked WRITE but lack destructiveHint annotation; mcp_list, api_endpoints, qa_list, qe_list lack readOnlyHint. LLMs cannot distinguish safe from risky operations without explicit hints.
Minimal parameter descriptions. 'api_plugin_slug' and 'quick_api_id' lack format guidance (e.g., expected length, character restrictions, examples of valid values). 'variables' in qa_run has no guidance on required vs optional keys or value types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
No error handling or recovery guidance. Tools do not document failure modes, retryability, or actionable error messages. An LLM receiving a 404 or timeout has no guidance on next steps.
Ambiguous tool naming for discovery. 'qa_list' (Quick APIs) vs 'qe_list' (Quick Execs) are not self-documenting. LLMs may conflate them or call both when only one is needed. Descriptions should clarify: 'Quick APIs are pre-configured HTTP calls; Quick Execs are shell commands.'