Local-first AI IDE with editor, Git, terminal, MCP, and human-in-the-loop coding agents.
Circle provides 19 tools with mostly complete schemas and descriptions. Naming follows verb_noun conventions (read_file, edit_file, delete_file, run_terminal_cmd). Most tools have 50-200 char descriptions and documented input parameters with types. However, several tools lack output schema documentation, some descriptions are generic, and error handling guidance is minimal. Tools like list_dir and glob_file_search have no visible input schema in the provided source. The ask_user tool has excellent schema depth, but most file/terminal tools are less detailed. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present despite clear risk classifications.
Ask the user multiple choice questions to gather information, clarify ambiguity, understand preferences, make decisions or offer them choices. Returns JSON: {"success":true,"answers":{...},"annotations":{...}} or {"success":false,"skipped":true} or {"success":false,"rejected":true,"feedback":"..."}.
Semantic code search powered by embeddings and sqlite-vec.
Delegate a self-contained sub-task to a specialized bounded agent (separate tool loop, same workspace).
Delete a file or directory from the filesystem. Supports undo through pending edits system.
Propose an edit to an existing file or create a new file. Uses a specialized model to apply changes intelligently.
Enter plan mode to explore the codebase and design an implementation approach before writing code.
Exit plan mode and request user approval of the plan before proceeding with implementation.
10 tools (list_dir, glob_file_search, grep, run_terminal_cmd, read_lints, update_memory, todo_write, get_skill_details, task_list, task_get, task_stop) have no visible input schema in source code.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present despite clear risk classifications (READ_ONLY, WRITE, DESTRUCTIVE). LLMs cannot infer safety properties without explicit hints.
Output schemas not documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Get details about available skills.
Search for files matching glob patterns.
Search for text patterns in files using grep.
List the contents of a directory.
Read the contents of a file from the filesystem.
Read linting results for files.
Run a terminal command in the workspace.
Get details about a specific sub-agent task.
List all active and completed sub-agent tasks.
Stop a running sub-agent task.
Write or update todo items.
Update the agent's memory with findings and context.
Destructive operations (delete_file, edit_file, run_terminal_cmd) lack confirmation/dry-run patterns. Agents can cause irreversible damage without safeguards.
Error handling descriptions missing. Tools do not guide LLMs on recovery actions (retry, ask user, abort). Generic descriptions like 'Run a terminal command' provide no error context.