Loading ToolBench
A control plane for agent-generated software delivery with MCP tool governance, workspace policies, and observability.
ControlKeel exposes 55 tools with minimal schema documentation. Only 1 tool (ck_tool_health) has visible input parameters with type definitions. The remaining 54 tools lack documented input schemas entirely, no JSON Schema, no parameter types, no descriptions. Tool descriptions are present but generic (10-50 chars), offering no guidance on when to use each tool or what it returns. No output schemas are documented. The codebase shows tool names defined in lib/controlkeel/mcp/tool_groups.ex but actual parameter and return schemas are not visible in the provided source. This violates the critical rule: 'If a tool has NO input schema at all: its schema score MUST be 0.' Per the rubric baseline, 100% of A+ tools have documented return types and 94% have docstrings, ControlKeel achieves neither at scale.
Attaches agents or resources to workspaces
Manages and tracks resource budgets for agent operations
Creates execution checkpoints
Lists available checkpoints
Restores from execution checkpoints
Provides contextual information for agent execution
Packages context information for agent operations
Provides copilot assistance for agent operations
54 of 55 tools have NO visible input schemas. Only ck_tool_health shows parameter definitions (session_id, session_limit with types). Remaining tools lack JSON Schema entirely, violating the critical rule: 'If a tool has NO input schema at all: its schema score MUST be 0.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Analyzes and suggests cost optimizations for agent operations
Delegates tasks to other agents or services
Provides deployment recommendations and guidance
Executes code within controlled governance boundaries
Indexes agent experiences for learning
Reads stored agent experiences
Searches agent experiences
Manages calls to external services
Analyzes and clusters failure patterns
Records security or quality findings
Finds files matching patterns
Searches file contents
Lists directory contents
Reads file contents
Creates git commits
Shows git differences
Reports git repository status
Manages agent goals and objectives
Loads resources for agent operations
Manages agent execution loops and iterations
Discovers available MCP tools and their capabilities
Archives agent memory records
Records information to agent memory
Searches agent memory for stored information
Provides observability and monitoring capabilities
Tracks outcomes and results of agent operations
Reports regression test results
Peeks at intermediate results without full commitment
Retrieves feedback from code reviews
Checks the status of submitted reviews
Submits code or changes for review
Rolls back agent operations to previous states
Routes agent requests to appropriate handlers or services
Manages agent sessions
Generates digests of agent sessions
Tracks skill development and evolution
Lists available skills
Loads a skill for use
Validates skill definitions
Manages individual tasks within agent workflows
Audits token usage and consumption patterns
Reports health and coverage metrics for tools
Traces execution packets through the system
Validates agent actions against governance policies and constraints
Manages workspace-level agent operations
Lists git worktrees
Switches between git worktrees
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns. Baseline: 100% of A+ tools have documented return types.
Tool descriptions are generic and under 50 characters (e.g., 'Validates agent actions against governance policies and constraints' for ck_validate). No guidance on WHEN to use each tool, WHAT it returns, or HOW it differs from similar tools. Baseline: average A+ tool description is 194 chars; descriptions under 20 chars score 0-20.
No parameter descriptions visible for 54 of 55 tools. Even ck_tool_health's parameters lack descriptions explaining what session_id and session_limit control. Baseline: 100% of A+ tool params have descriptions.
No error handling guidance documented. Tools like ck_execute_code (WRITE risk) and ck_git_commit (WRITE risk) provide no recovery hints, retryability classification, or actionable error messages. Agents cannot self-correct on failure.