fix agent analysis command routing: add per-tool slash commands and ensure 'text extraction' maps to standalone OCR

[dvschultz]

Summary

Agent routes requests incorrectly (e.g., 'text extraction' triggers classification) and /analyze runs everything unexpectedly. Add explicit per-tool slash commands, require confirmation for run-all, and fix intent mapping so OCR can run standalone.

Source

• File: GMT20260512-000826_Recording.m4a
• Path: /Volumes/Lexar/2026 DADA ZOOM/Week 5/GMT20260512-000826_Recording.m4a
• Transcript time: 00:57:08 - 01:00:25

Evidence

Need individual analysis slash commands; /analyze runs everything; 'text extraction' maps wrong and OCR not standalone (GMT20260512-000826_Recording.m4a 3428–3510, 3553–3625)

Acceptance Criteria

• Chat /help lists individual analysis commands with short descriptions and required/optional args
• Running /analyze without args prompts for confirmation or requires explicit selection of tools
• Natural-language requests for 'text extraction'/'text on screen' map to OCR tool (not classification)
• OCR can be invoked as a standalone operation via slash command (or agent explains prerequisites and offers to run them)

Implementation Notes

Introduce command router with tool registry; add intent-to-tool mapping tests for OCR vs classification.

Extraction confidence: 0.83