A terminal coding agent that helps with software engineering tasks in a repository using RAG search, LSP tools, shell execution, and file operations
Claus provides 12 tools with basic schemas and descriptions, but falls short of production quality. All tools have descriptions (10-100 chars, mostly adequate) and input schemas with type definitions. However, parameter descriptions are sparse or missing context. No output schemas documented. Error handling is absent, tools return raw errors without recovery guidance. Tool names follow verb_noun convention (read_file, write_file, edit_file, list_dir, shell, search_text, lsp_*). Critical gaps: no parameter validation hints, no enum constraints for constrained inputs, no idempotency markers, no permission gates for destructive operations (write_file, edit_file, shell, dispatch_agent). The shell tool is particularly risky, no sandboxing hints, no command validation guidance. RAG and LSP tools lack output schema documentation. Composition is reasonable (tools chain via file paths and line numbers), but missing pagination for rag_search and list_dir.
Delegate a self-contained task to a sub-agent with its own context and the same tools. Give it a complete task description; it returns only its final report.
Edit a file by replacing a range of lines
List the contents of a directory
Find the definition of a symbol using LSP
Get diagnostics for a file using LSP
Get hover information for a symbol using LSP
Find all references to a symbol using LSP
Search the project using semantic RAG search to find relevant code by meaning
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what does rag_search return? what fields does lsp_hover provide?). Violates pattern:tool and pattern:response-shaper.
Destructive/irreversible tools (write_file, edit_file, shell, dispatch_agent) lack permission gates, dry-run support, or confirmation patterns. No guidance on error recovery or rollback. Violates pattern:permission-gate and pattern:confirmation-request.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Read the contents of a file
Search for text patterns in files
Execute a shell command and return its output
Write contents to a file, creating it if it does not exist
shell tool description is minimal ('Execute a shell command and return its output'). No guidance on sandboxing, command validation, injection risks, or what commands are allowed. LLMs may pass arbitrary/dangerous commands. Violates pattern:tool-description and pattern:tool-gateway.
LSP tools (lsp_definition, lsp_references, lsp_hover, lsp_diagnostics) lack output schema documentation and error handling. What does lsp_hover return? What fields? What if LSP server is unavailable? Violates pattern:tool and pattern:recovery-guide.
rag_search lacks pagination (no limit default, no offset/cursor, no total count). Large result sets will blow context window. Violates pattern:paginated-result.