Loading ToolBench
Persistent, cross-client project memory for AI coding agents and MCP-compatible tools. Backed by Supabase or SQLite with hybrid semantic/lexical search, session management, and deployment tracking.
22 tools with Spanish descriptions and basic schemas. All tools have input schemas with typed parameters and descriptions, but descriptions are often generic or lack actionable context. No output schemas documented. Tool names follow verb_noun pattern (resolve_project, create_project, etc.) but lack clarity on when to use similar tools (e.g., resolve_project vs create_project). No error handling guidance, no recovery hints, no parameter constraints beyond enums on 3 tools. Security concerns: no evidence of permission gates, audit trails, or secret injection patterns. Composition issues: tools like capture_project_memory and load_unified_context combine multiple concerns. Spanish descriptions reduce LLM clarity in English-first environments.
Registra advertencias inteligentes.
Aplica politicas de retencion y archivo.
Guarda automaticamente decisiones, tareas, checkpoints, archivos y estado en una sola llamada.
Crea un proyecto con metadata de repo y workspace.
Crea una sesion y guarda contexto git.
Finaliza una sesion y genera resumen automatico.
Spanish descriptions reduce LLM clarity in English-first environments. Descriptions like 'Resuelve o crea un proyecto automaticamente' and 'Crea un proyecto con metadata de repo y workspace' lack actionable context for English-speaking LLMs. Descriptions should be 50-200 chars in English, explaining WHAT, WHEN, and WHY.
No output schemas documented. Tools return data but LLMs cannot plan downstream calls without knowing response structure. E.g., list_projects, search_semantic_memory, load_unified_context lack documented return types. This violates pattern:response-shaper and forces LLMs to guess field names.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | <=2025-11-25 | v2 |
Exporta memoria a JSON o Markdown.
Lista advertencias activas.
Resume uso por interfaz.
Devuelve la linea de tiempo del proyecto.
Importa memoria desde JSON o diccionario.
Lista proyectos del owner o workspace actual.
Carga contexto persistente optimizado.
Resuelve o crea un proyecto automaticamente.
Devuelve un resumen listo para retomar el proyecto.
Guarda checkpoints de arquitectura y estado.
Guarda decisiones compartidas.
Guarda memoria por archivo y relaciones.
Guarda prompts utiles y patrones de respuesta.
Busca memoria con ranking lexical y semantico combinado.
Sincroniza el estado de sesion.
Crea o actualiza tareas del proyecto.
No error handling guidance. Tools like apply_retention_policy (DESTRUCTIVE risk) and import_memory_bundle lack recovery hints. No indication of retryable vs fatal errors, no actionable error messages. Violates pattern:recovery-guide.
Destructive tool (apply_retention_policy) lacks confirmation/dry-run pattern. No evidence of permission gates or audit trails. Violates pattern:confirmation-request and pattern:audit-trail. A dry_run parameter exists but no guidance on what happens without it.
Overlapping tool responsibilities. resolve_project (creates if missing) and create_project both create projects. load_unified_context and capture_project_memory both aggregate project state. LLMs will struggle to choose between similar tools. Violates pattern:tool (single responsibility).