Loading ToolBench
Built-in skill definitions for the Zibby agent-workflow framework, providing MCP tools for file operations, integrations (Figma, PingCode, GitHub, Slack, Jira, Linear, etc.), memory, reporting, and more.
Mixed quality across 21 tools. Core tools (read_file, write_file, list_directory, run_command) have clear verb-noun naming and basic descriptions, but lack detailed parameter constraints and output schemas. Figma tools are well-named and moderately described but missing output schema documentation. PingCode tools have sparse parameter descriptions (many null). No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Error handling guidance is absent across all tools. Average description length ~120 chars (below 194-char baseline for A+ tools). Parameter descriptions often missing or generic.
Returns whether the current user has a valid PingCode authorization, and when it was granted / expires.
Create a new ticket (产品工单) as the authenticated user. Call list_products and list_ticket_types first to get the required IDs. Always show the user the html_url from the response.
List all comments on a file (id, message, author, resolved status, position).
Fetch a Figma file by key or URL. Returns the document node tree, components and styles. Large files are big — prefer `depth` (1-2) or figma_get_file_nodes for a subtree.
Fetch only the given nodes (and subtrees) — much smaller than figma_get_file. Node ids accept URL form `1-23` or API form `1:23`.
Return the saved version history (version id, label, created_at, user). Use a version id with figma_get_file.
Output schemas undocumented. No tool returns documented field types, structures, or pagination info. LLMs cannot plan downstream calls or extract required data.
PingCode tools have null parameter descriptions (product_id, type_id, state_id, priority_id, etc.). LLMs cannot infer what these fields control or what values are valid.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. write_file, run_command, create_ticket, update_ticket, figma_post_comment are destructive but unmarked.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Return a map of imageRef → download URL for every image fill used in the file.
Render the given nodes to PNG/JPG/SVG/PDF and return temporary URLs (valid ~14 days). Use for previews/exports.
Dev-Mode-style inspection computed from the REST tree: for the given nodes and their children, returns size, position, colors (hex), fonts, spacing/auto-layout, radius, strokes and effects — a compact spec for building UI. Node ids accept `1-23` or `1:23`.
Add a comment to a file. Optionally reply to a comment (comment_id) or pin it to a node. Needs a PAT with comment-write scope.
Returns the current Figma user (id, handle, email, img_url) — a quick way to confirm the token works.
List files and directories in a path. Returns names with type indicators (/ for dirs).
List all Ship products visible to the authenticated user. Returns id, name per product.
List available ticket types for the given product. Required before creating a ticket.
List tickets visible to the authenticated user with optional filters. All filters optional.
Open a URL in the user's default browser. Use for OAuth flows, documentation, integration setup pages.
Read the contents of a file. Returns the text content.
Run a shell command and return its output. Use for grep, git, npm, etc.
Patch an existing ticket. Only include fields you want to change.
Wait for N seconds. Use this for async operations (tests, builds, deploys) — wait, then check status again.
Write content to a file. Creates parent directories if needed.
No error handling guidance. Tools lack recovery hints (e.g., 'If file not found, check path with list_directory'). Agents cannot self-correct on failures.
Parameter constraints missing. run_command lacks cwd validation; figma_get_file lacks depth range; list_tickets lacks page_size bounds. LLMs can pass invalid values.