Scoring was not performed
Tool descriptions too brief (10-40 chars), lack WHAT/WHEN/WHEN-NOT context for LLM selection. 'Search for files using glob patterns' does not explain when to call search_files vs read_directory or what field structure to expect in results.
Parameter descriptions are generic or missing. 'pattern' (search_files) has only 'Glob pattern to search for (e.g., "*.ts", "**/*.json")', does not explain max glob depth, whether recursive by default, case sensitivity, or symlink handling. LLMs cannot infer these constraints from examples alone.
No output schema documented for any tool. LLMs cannot plan downstream operations or extract the right fields. E.g., what fields does search_files return? Is it {name, path, size, type}? Does it include metadata like modified time? Without this, agents cannot chain calls reliably.
No pagination or result limit controls visible in schemas. search_files accepts maxResults (default 50), but no total_count or next_cursor returned. If glob matches 1000+ files, agent has no way to iterate, either gets silent truncation or context bloat.
No error handling guidance. What happens if read_file_content hits maxSize limit? Does it return partial content or error? Does the agent retry, ask the user, or abort? No recovery paths defined.
path traversal security not addressed in descriptions. read_file_content and read_directory accept 'path' parameter, no mention of rootPath enforcement, symlink restrictions, or bounds checking. LLMs could be tricked into reading /etc/passwd or parent directories.
maxSize parameter (read_file_content) has no description explaining units (bytes?), default value, or what happens when exceeded. LLMs may pass nonsensical values like maxSize=1 or interpret the parameter as characters instead of bytes.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 23 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | - | v1 |