An open-source agent platform with desktop shell, web interface, and sidecar services for AI-powered automation and coding assistance
Lume exposes 8 tools with critical definition gaps. Tool names lack action verbs (bash, memory.promoteglobal, automation_set, write, edit, multiedit, notebookedit). Descriptions are present but minimal (10-50 chars), providing no context on WHEN to use each tool or what happens on failure. Input schemas visible for only 3 tools (bash, write, edit, multiedit, notebookedit show file_path/edits/notebook_path); memory and automation_set tools have no visible schema. No output schemas documented. No parameter descriptions beyond the field name. No error handling guidance. No idempotency or retry semantics declared despite destructive operations (bash, write, edit). Tool names like 'memory.promoteglobal' and 'memory.rejectglobalcandidate' use dot notation and lack clarity on what 'promote' and 'reject' mean in context. No enum constraints on parameters. Descriptions do not explain prerequisites, side effects, or recovery paths.
Modify automation tasks affecting future scheduled execution
Execute bash commands with safety guardrails and analysis
Edit file content with boundary checking
Promote memory to global scope affecting cross-workspace memory
Reject global memory candidate affecting cross-workspace memory
Edit multiple files with boundary checking
Edit notebook content with boundary checking
No input schemas visible for memory.promoteglobal, memory.rejectglobalcandidate, automation_set. Cannot verify parameter types or constraints.
Tool names lack action verbs. 'bash', 'write', 'edit' are weak; 'memory.promoteglobal' uses dot notation and is unclear. Should be 'execute_bash', 'write_file', 'edit_file', 'promote_memory_to_global'.
Descriptions are too short (10-50 chars) and lack context. No explanation of WHEN to use each tool, what side effects occur, or how to recover from errors. Baseline for A+ tools is 50-200 chars with clear intent.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Write file content with boundary checking
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. Baseline: 100% of A+ tools have documented return types.
Destructive tools (bash, write, edit, multiedit, notebookedit) have no error handling guidance, confirmation steps, or dry-run support. No indication of idempotency or retry safety.