Loading ToolBench
Configuration repository and maintenance scripts for OpenCode CLI, including agent definitions, team configurations, profiles, and deployment/health management utilities
This bash-based MCP server has significant definition quality gaps. Tool descriptions are present but lack LLM-optimization guidance (e.g., 'maintain' is 127 chars but vague on when to use it vs. 'cleanup'). Input schemas are visible in the source but lack formal JSON Schema structure, parameters are documented as bash flags (--check, --clean) rather than typed schema objects. No output schemas are documented. Error handling is minimal (bash exit codes only). Tool naming is reasonable (verb-noun style) but some tools conflate multiple concerns (e.g., 'diagnose' does signal collection, AI diagnosis, and fix application in one tool). No parameter descriptions explain constraints, ranges, or dependencies. Security: credentials appear to be handled via environment (OC_SESSIONS_DIR, OC_BACKUP_ROOT) but not explicitly validated in code. Composition: tools like 'maintain', 'cleanup', and 'doctor' overlap significantly in scope.
Reconcile the environment back to the known-good setup: validate required files, config symlink, plugin pin, cache pruning, leftover copies, backups, and repo cruft removal
Pre-deployment gate + runtime protection: credit balance checks, model health probes, rate limit verification, git tree validation, signature verification, deployment locking, and quarantine mode for cost-saving
Deep, AI-assisted diagnostician: collects live signals from opencode, plugin cache, default-agent resolution, and OpenRouter (key usage/limits, model routing), then requests AI diagnosis and proposes fixes
Full OpenConfig readiness check: CLI, config link, signature, plugin cache, LSP, formatters, keys, prompts, colors, models, MCP, permissions, concurrency, teams, OmO footguns, content-aware skills, terminal, telemetry, runtime logs, and external-source hardening
Export the shared, secret-free OpenConfig runtime for Linux T3: generates manifest with SHA256 hashes of config files, agents, prompts, profiles, skills, and teams
System maintenance: clean stale sessions, prune caches, check disk usage, and collect plan strategies for optimization
No formal JSON Schema for tool inputs. Parameters documented as bash flags (--check, --clean) rather than typed schema objects with constraints, ranges, or enums. LLMs cannot infer valid parameter combinations or constraints.
No output schemas documented. Tools return bash output (text, JSON from Python) but LLMs have no specification of return structure, field types, or pagination. Agents cannot plan downstream calls reliably.
Tool 'diagnose' conflates three concerns: signal collection, AI-assisted diagnosis, and fix application (--fix, --agent-fix). Should split into separate tools (collect_diagnostics, request_ai_diagnosis, apply_fixes) for composability.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | 2026-07-28+ | v2 |
OpenRouter admin dashboard + live health analysis for account overview, credits, keys, spend, rate limits, model routing, and health checks
Descriptions lack LLM-optimization guidance. 'maintain' (127 chars) and 'cleanup' (156 chars) both manage stale files but distinction is unclear. No 'when to use' guidance. Descriptions should be 50-200 chars with clear differentiation.
No error handling guidance. Bash scripts exit with codes but provide no recovery hints. E.g., if 'openrouter-admin health' fails, LLM has no guidance on whether to retry, check credentials, or escalate.