DreamCleanr defines 5 tools with complete input schemas and tool annotations (readOnlyHint present). All tools have descriptions (10 - 200 chars, within baseline 34 - 392 range). However, parameter descriptions are entirely absent, the schema shows enums and types but no explanation of what each parameter controls. Output schemas are undocumented; the code shows tool_scan returns a summary dict but the structure is not formally specified. Error handling is minimal; no recovery guidance or actionable error messages visible. Tool names are clear (scan, clean_preview, report_render, schedule_status, schedule_preview) but lack verb-noun consistency (scan vs. clean_preview vs. report_render). The report_render tool is marked READ_ONLY in annotations but performs a WRITE operation (html_out file output), creating a semantic mismatch.

• Add a description to every parameter. For 'mode', explain: 'Cleanup aggressiveness: safe (minimal risk), balanced (recommended), or max (aggressive). Default: balanced.' For 'scope', explain: 'What to clean: all (processes + storage), processes only, or storage only. Default: all.'
• Document output schemas formally. For scan, specify: 'Returns {process_summary: {family: {state, recommended_action, stale}}, detectors: {key: {title, status, total_bytes, path_count, cleanup_ready, safety_state, active_project_count}}}'.
• Fix report_render annotation: change readOnlyHint to false or add destructiveHint=true, since it writes html_out to disk. Ensure annotations match actual behavior.
• Add error handling with recovery guidance. E.g., 'If mode is invalid, return: {error: "Invalid mode. Must be safe, balanced, or max.", recovery: "Retry with a valid mode."}' for each tool.
• Standardize naming to verb_noun: rename report_render → render_report, schedule_status → get_schedule_status, schedule_preview → preview_schedule. This improves LLM parsing and consistency.