A web-based agent interface and orchestration platform with support for subagents, project previews, visualizations, and delegated task execution
Hlid exposes 17 tools with basic naming and descriptions, but critical gaps undermine production readiness. Tool names follow verb_noun convention (delegate_, list_, inspect_, etc.), which is positive. However, descriptions are uniformly terse (15 - 40 chars), falling well below the 50 - 200 char baseline for LLM-optimized docs. Input schemas are present but minimal: most parameters lack type constraints, enums, or validation guidance. No output schemas are documented. Error handling is absent, tools provide no recovery guidance, retryability classification, or actionable error messages. The delegation and preview tools (delegate_hlid_agent, steer_hlid_agent, control_project_preview) lack parameter constraints that would prevent hallucinated values. No tool declares permissions, audit trails, or security boundaries. Composition is reasonable (each tool has one responsibility), but chaining is fragile, responses do not document what IDs/references downstream tools need.
Request cancellation of a running child Hlid agent
Capture a screenshot or state snapshot of an active project preview
Capture a screenshot of the active web browser session
Clean up the worktree associated with Hlid agent orchestration
Send control commands (clicks, input, navigation) to an active project preview
Send control commands to the active web browser session
Delegate a task to a child Hlid agent with specified task description and optional provider
Descriptions uniformly terse (15 - 40 chars), far below 50 - 200 char LLM-optimized baseline. Examples: 'Delegate a task to a child Hlid agent with specified task description and optional provider' (88 chars, but lacks WHEN to use, prerequisites, or return value guidance).
Input schemas present but lack type constraints, enums, and validation guidance. Parameters like 'instruction' (steer_hlid_agent), 'action' (control_project_preview), and 'provider' (delegate_hlid_agent) are free-form strings with no enum or pattern constraints, inviting hallucinated values.
No output schemas documented. Tools return results but LLMs cannot plan downstream calls or extract required IDs. Example: list_hlid_agents returns agent list, but response structure (fields, pagination, total count) is undocumented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Generate media (images) using specified provider with prompt and optional parameters
Create an interactive HTML visualization with optional title and attachment storage
Inspect the current state and details of a child Hlid agent
List all child Hlid agents currently managed by this orchestration context
Open a web browser session for capturing and controlling web content
Resume execution of a paused child Hlid agent with optional instruction
Start a project preview session for capturing and controlling project state
Start recording a web browser session for playback and analysis
Send steering instructions to a running child Hlid agent
Wait for a child Hlid agent to complete its task
No error handling guidance. Tools provide no recovery hints, retryability classification, or actionable error messages. An LLM receiving 'Agent not found' has no next step; should be 'Agent not found. Call list_hlid_agents() to see available agents.'
No security declarations. Tools lack permission gates, audit trail documentation, or scope declarations. Destructive tools (cancel_hlid_agent, cleanup_hlid_worktree) should declare required permissions and support dry-run or confirmation patterns.