Loading ToolBench
Migrate, back up, and version your AI coding assistant environment (Claude Code / Codex) — MCP servers, Skills, Agents, Plugins, Hooks, and memory — as a portable, secret-isolated package.
AgentDock CLI is a STDIO-only tool management utility with 9 commands. Tool definitions are inferred from TypeScript command files rather than explicitly registered via MCP protocol. All tools have descriptions (10-200 chars, baseline 194), but input schemas lack formal JSON Schema type declarations, parameters are documented as strings with descriptions but no explicit 'type' or 'enum' constraints. No output schemas are documented. Error handling is minimal; no recovery guidance or actionable error messages visible. Tools combine multiple concerns (e.g., 'scan' generates manifests AND isolates secrets; 'doctor' diagnoses AND checks for leaks). Parameter descriptions are present but lack format constraints, ranges, or validation rules. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) detected.
Manage AgentDock configuration (list, get, set). Supports keys: agent, out, env, lang.
Compare an AgentDock package against a target environment to show what would be installed.
Diagnose the health of a Claude Code or Codex environment. Checks configuration validity, migratable content, secret leaks, and run-state isolation. Supports live mode, scan artifacts, and install packages.
Export a manifest (agentdock.yml) into an installable package with payload, snapshot, and install plan. Supports exporting from scan artifacts via --from-scan.
Install an AgentDock package into a target environment (Claude Code or Codex home directory). Supports dry-run mode and conflict detection.
List all MCP servers, Skills, Agents, Plugins, Hooks, Memory, and Settings in a scan artifact or install package.
No formal JSON Schema type declarations for parameters. All inputs documented as strings with descriptions, but no 'type', 'enum', 'minimum', 'maximum', or 'pattern' constraints visible in source. LLMs cannot validate inputs or select from constrained options.
No output schemas documented. Tools return results (manifests, reports, diffs) but LLMs have no visibility into response structure, field names, or types. Prevents downstream tool chaining and forces LLMs to parse unstructured output.
Tool definitions inferred from command files, not explicitly registered via MCP protocol. No visible tool registration with names, descriptions, and schemas in MCP-compliant format.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
Scan the local AI coding assistant environment (Claude Code or Codex) and generate a manifest of MCP servers, Skills, Agents, Plugins, Hooks, and memory with secret isolation.
Remove an AgentDock package from a target environment. Supports dry-run mode and force flag.
Validate an agentdock.yml manifest against the schema.
No error handling guidance. Tools document risks (READ_ONLY, WRITE, DESTRUCTIVE) but provide no recovery hints, error classification, or actionable messages. LLMs cannot self-correct on failures.
Multiple concerns per tool. 'scan' generates manifests AND isolates secrets. 'doctor' diagnoses AND checks for leaks. 'config' lists, gets, and sets via subcommand. Violates single-responsibility principle; prevents independent composition.