Two tools with detailed parameter schemas but significant quality gaps. Both tools have descriptions (good), but descriptions are overly technical and lack actionable guidance for LLM selection. Parameter schemas are comprehensive but lack type information for many fields (channels array has no item type; several numeric thresholds lack min/max bounds). No output schema documented. No error handling guidance. Tool names lack clear verb-noun structure ('assert_pixels' vs 'compare_pixels_in_region' would be clearer). Parameters like 'match_by' and 'command' lack enum constraints despite accepting specific values.

• Document output schema for both tools. Specify: {success: boolean, capture_file: string, operation: string, result: object, error?: string}. Include what 'result' contains (pixel_diff_metrics, state_diff_ratio, etc.).
• Convert 'match_by' parameter to enum: ["event_id", "index", "name"]. Add bounds to numeric thresholds: max_mean_abs_diff (0-255), min_psnr (0-100), max_failed_pixel_ratio (0-1.0).
• Rewrite descriptions to answer: What does this tool verify? When should I call it instead of the other? What does success look like? E.g., 'Verify pixel differences in a render region stay within acceptable thresholds. Use this to detect visual regressions. Returns pass/fail verdict with metrics.'
• Add error handling guidance: 'If timeout_secs exceeded, the capture file may be corrupted or RenderDoc unavailable. Retry with a larger timeout or check RenderDoc installation.' Document what each threshold violation means.
• Declare 'channels' array item type as string. Add description to 'command' parameter explaining when to override the default RenderDoc path (e.g., custom build location).