Techne exposes 48 tools across filesystem and language-specific operations. Critical gaps: NO visible input/output schemas in source code (internal/app/app.go, pkg/fs/tools.go, pkg/lang/go/tools.go show tool registration but no schema definitions). Tool descriptions are absent or minimal, cannot verify they meet the 10-1024 character baseline. Parameter documentation is not visible in the provided code. Tool names follow verb_noun convention (fs.read, fs.write, lang.go.rename) which is good, but without schemas and descriptions, tools cannot guide LLM selection. The codebase uses github.com/google/jsonschema-go for schema generation from Go struct tags, but the actual struct definitions and their tag annotations are not shown, making it impossible to verify schema quality. Per HARD SCORING RULES: if schemas are not visible, schema score must be 0; if descriptions are not visible, description score must be 0.
No input/output schemas visible in source code. Tool.go shows schema generation via jsonschema-go struct tags, but actual struct definitions with tag annotations are not provided. Cannot verify parameter types, constraints, or output field documentation.
Tool descriptions are not visible in provided source code. Cannot verify they meet the 10-1024 character baseline or explain WHEN to use each tool vs alternatives. LLMs cannot select tools without descriptions.
Destructive operations (fs.delete, lang.go.delete_file) lack visible confirmation/dry-run patterns. No evidence of error handling that guides recovery or explains consequences to LLMs.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 32 | 2026-07-28+ | v2 |
Multi-step refactoring tools (extract_function, extract_interface, move_package) lack visible composition guidance. No evidence that tool outputs include IDs/references needed by downstream tools.
No visible error handling patterns. Cannot verify tools return actionable error messages that guide LLM recovery (e.g., 'File not found. Try fs.find() first.').