ResonantOS add-on that wraps the vendored grocery_flywheel package and exposes grocery planning, dietary analysis, sourcing research, replenishment analysis, and dashboard rendering via HTTP-JSON on 127.0.0.1:4899
Server exposes 5 tools with complete input schemas and parameter descriptions. Naming follows verb_noun convention (analyze, dietary, sourcing, plan, dashboard). However, descriptions are generic and lack actionable context for LLM selection. Parameter validation is thorough server-side (state_json length, objective enum, filename regex), but descriptions do not explain WHEN to use each tool or what distinguishes them. Output schemas are not documented. Error handling returns HTTP status codes but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles (READ_ONLY vs WRITE).
Analyze replenishment state using the vendored MCP tool handler analyze_replenishment_state
Render and persist dashboard HTML under var/dashboards/ with home-path redaction and single-flight concurrency control
Evaluate dietary constraints using the vendored MCP tool handler evaluate_dietary
Plan next cart with approval boundary enforcement using the vendored MCP tool handler plan_next_cart
Summarize sourcing research using the vendored MCP tool handler summarize_sourcing_research
Tool descriptions lack actionable context. All 5 tools describe WHAT they do (via vendored handler names) but not WHEN to use them or how they differ. E.g., 'Analyze replenishment state' vs 'Evaluate dietary constraints', an LLM cannot infer when to pick one over the other without domain knowledge.
No output schema documentation. Tools return results from vendored handlers, but the response structure is not documented. LLMs cannot plan downstream calls or extract required fields without knowing what fields are returned.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
No tool annotations despite clear risk profiles. plan and dashboard are WRITE operations with side effects (approval boundary, file persistence), but lack destructiveHint/idempotentHint annotations. Analyze, dietary, sourcing are READ_ONLY but lack readOnlyHint. LLMs cannot reason about retry safety or side effects.
Error responses lack recovery guidance. Code returns HTTP 400/409 with error strings, but does not guide the LLM on next steps. E.g., 'unknown objective' should suggest valid objectives; 'dashboard render in progress' should explain retry behavior.
Parameter descriptions are minimal. 'objective' is described as 'Optional objective from OBJECTIVES enum' but does not explain what objectives are available or how they affect tool behavior. 'mode' in plan tool lacks explanation of pickup vs delivery vs in_person trade-offs.