Loading ToolBench
A local read-only report helper for AI coding agents. Provides project-specific report rules, configuration management, and safety requirements without accessing project code, Git diffs, prompts, or secrets.
JuTell's 5 tools are all read-only diagnostic/configuration getters with clear, domain-specific descriptions (100-200 chars each). All tools have empty input schemas ({}), which is appropriate for parameter-less operations. Descriptions are well-written and explain WHEN to use each tool vs. alternatives (e.g., 'Use only when asked how JuTell is configured; get_beginner_report_rules already includes these limits'). However, tools lack output schema documentation, the response structure is not formally specified. No error handling guidance is present. Tool naming follows verb_noun pattern (get_*) correctly. The domain is narrow (JuTell configuration introspection), so composition and chaining are minimal concerns.
Diagnostic only: list which JuTell report sections are on/off and what each safely omits when off. Use only when you need to explain JuTell's own configuration to the user; get_beginner_report_rules already returns the active sections needed to write a report, so call this instead of that only for that narrower question.
The primary report-writing tool: returns this project's active report rules (which sections to include, length/voice limits, the evidence/status/diff rules, safety requirements) in one call. Call this once, right before writing the final owner-facing report for a completed task — it replaces reading report-format.md, risk-level-guide.md, and related Skill reference files by hand. Do not call it for non-report answers (plain questions, explanations, plans), and do not call it more than once per task just to double-check.
Diagnostic only: return whether JuTell is configured and enabled for this project. Use when JuTell setup itself looks wrong (e.g. reports seem to ignore .jutell.json), not as part of a normal report — get_beginner_report_rules already includes what a report needs. Does not read project code or Git.
Diagnostic only: return the report Profile (minimal/balanced/learning/detailed) and length/file/glossary limits. Use only when asked how JuTell is configured; get_beginner_report_rules already includes these limits for normal report writing.
No output schemas documented for any tool. LLMs cannot infer response structure, field types, or downstream chaining requirements.
No error handling guidance. Tools do not document failure modes (e.g., missing .jutell.json, invalid config) or recovery steps.
Input schemas are empty ({}). While correct for parameter-less tools, the MCP SDK should explicitly declare inputSchema with type 'object' and properties: {} for clarity.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Return the fixed list of items (failures, unresolved risks, secrets exposure, data loss, out-of-scope changes) that a report must never omit, even when the active Profile would otherwise skip them. Use alongside get_beginner_report_rules only when a report involves a failure, risk, or safety-relevant change — most routine reports do not need a separate call for this.