Loading ToolBench
MCP server plugin for ShipToday Forge that provides workflow orchestration, session observation, and state management hooks for Claude Code and Codex environments
Critical gaps across all dimensions. Tool definitions lack proper schemas, descriptions are vague or missing context, and naming is inconsistent. 5 of 7 tools have no visible input schema validation in source. Descriptions are generic and fail to explain WHEN to use each tool or what it returns. No error handling guidance. Tool names mix underscores and double-underscores inconsistently (forge__start_workflow vs forge-autopilot). The server appears to be a Claude plugin wrapper rather than a proper MCP server with well-defined tool contracts.
Skill that orchestrates SDLC workflows (planning, implementation, review, status) for tracked work items and provides routing for epic key references and workflow continuations
Terminates an active Forge workflow with a meaningful reason when the user has redirected to unrelated work
Retrieves the current state of an active Forge workflow for display and decision checkpoints
Initiates a Forge workflow for SDLC actions (planning, implementation, review, status) on tracked work items
Updates Forge workflow state with engineering-time deltas, skill invocations, and checkpoint data for session tracking
Codex/Claude Code hook event handler that normalizes tool invocations and validates Forge workflow tool calls with literal input parsing
No visible input schema validation or JSON Schema definitions in source code. Tool definitions show parameter names and types in comments only, not in machine-readable schema format. Cannot verify LLM will receive proper type constraints.
Descriptions lack actionable context. 'Initiates a Forge workflow' does not explain WHEN to call this vs other tools, what prerequisites exist, or what the return value contains. Descriptions average ~60 chars; baseline is 194 chars for A+ tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Passive session observation workflow that tracks engineering time and session metadata for linked/logged sessions, with optional user prompting for snoozed sessions
Inconsistent naming convention. Tools use both double-underscore (forge__start_workflow) and hyphen (forge-autopilot) separators. No consistent verb_noun pattern. 'session_observer' and 'functions.exec' do not start with action verbs (should be 'observe_session', 'execute_code').
No output schema documentation. Tool descriptions do not specify what fields are returned, their types, or how downstream tools should consume the results. Agents cannot plan multi-step workflows without knowing response structure.
No error handling guidance. Tool descriptions do not explain what errors can occur, whether they are retryable, or what the LLM should do next. Code in tool-event.cjs validates literal calls but provides no recovery paths.