Three tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. pp_audit has a reasonable description (145 chars) and input schema with three parameters, but lacks output schema documentation. pp_trend and pp_fix have minimal descriptions (under 80 chars) and sparse parameter details. No parameter descriptions for 'count' in pp_trend or 'dryRun' in pp_fix. Error handling and recovery guidance absent. Tool names follow verb_noun convention (audit, trend, fix) but lack clarity on what 'trend' returns or how 'fix' differs from 'audit'. No tool annotations (readOnlyHint/destructiveHint) despite pp_fix being destructive. Composition is reasonable, three separate concerns, but output chaining unclear (e.g., does pp_audit return IDs that pp_fix accepts?).

• Document output schemas for all three tools. Example for pp_audit: 'Returns {scores: {disk: number, memory: number, ...}, overall: number, recommendations: string[]}'. This enables LLMs to extract and chain results.
• Add descriptions to all parameters. pp_trend.count: 'Number of historical entries to display (1-100, default 10)'. pp_fix.dryRun: 'If true, show what would be fixed without executing changes (default false)'.
• Expand pp_trend and pp_fix descriptions to 150+ chars. Include WHEN to call: 'pp_trend: Call after pp_audit to see if scores are improving or degrading over time.' 'pp_fix: Call after pp_audit identifies issues; use dryRun=true first to preview changes.'
• Add tool annotations in the TOOLS array. Mark pp_audit and pp_trend with readOnlyHint: true. Mark pp_fix with destructiveHint: true and idempotentHint: false (since dryRun=true is idempotent, but dryRun=false is not).
• Add error handling guidance. Document what pp_fix returns on partial success (e.g., 'cleaned npm cache (freed 500MB), but temp files locked, retry later'). Include recovery hints: 'If pp_fix fails, check disk space or running processes.'