A configurable local agent harness for model and API tool calls, supporting built-in tools, API tools, MCP servers, and skills.
Tide has 18 tools with significant definition gaps. Most tools lack visible input schemas in the provided source (only skill and install_skill show complete schemas). Descriptions vary widely: skill and install_skill are well-written (100+ chars, actionable), but 16 builtin tools (copy_path, create_directory, delete_path, list_files, glob, grep, move_path, read_file, write_file, replace_in_file, run_command, get_command_output, web_fetch, web_search, todo_write, spawn_agent) have no visible schema or description in the source code provided. Per HARD SCORING RULES, tools without visible schemas score 0 on schema dimension. This drags the average significantly. The two well-defined tools (skill, install_skill) score ~75 each; the 16 undefined tools score ~20 each (inferred existence, no visible definition). Average: (75+75+20*16)/18 ≈ 42.
Install a skill into the local skills directory from a local folder path or a git URL (https://..., git@..., or github:owner/repo). The source must contain a SKILL.md. The skill becomes available immediately after installing — no restart needed.
16 of 18 tools lack visible input schemas and descriptions in source code. Only skill and install_skill have complete, documented schemas. Builtin tools (copy_path, create_directory, delete_path, list_files, glob, grep, move_path, read_file, write_file, replace_in_file, run_command, get_command_output, web_fetch, web_search, todo_write, spawn_agent) are defined in .js files not provided, making schema validation impossible.
No error handling guidance visible. Tools like delete_path and run_command are destructive/irreversible but lack recovery hints. LLMs cannot determine if errors are retryable or fatal.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 27 | 2026-07-28+ | v2 |
Load and follow an installed skill's instructions. Use when a request matches one of the available skills below. Calling this returns the skill's full instructions; then carry them out.
Destructive tools (delete_path, run_command, spawn_agent) lack confirmation or dry-run patterns. Agents can irreversibly delete files or spawn processes without safeguards.
No documented output schemas for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., does read_file return {content, size, encoding}? Does list_files return {files, total_count}?).
Parameter naming inconsistency and ambiguity. Tools accept 'path', 'source', 'name' without clear type hints (file path vs directory path vs URL). No enum constraints visible for tools like run_command (shell vs bash vs sh?).