What problem does it solve? When AI coding agents repeatedly produce plans that get denied, the feedback history contains valuable patterns that are hard to see manually. This Skill reads your entire Plannotator plan archive (or Claude Code ExitPlanMode logs as a fallback), extracts every denial reason, and turns them into a taxonomy, recurring-phrase analysis, and concrete prompt instructions that prevent future denials. ## Core Features & Use Cases - Map-reduce denial analysis: Parallel extraction agents read every denied plan file, then reduction agents cluster feedback into a taxonomy with real counts, quotes, and evolution over time. - HTML dashboard report: Produces a self-contained, editorial-style HTML report with 7 sections covering denial taxonomy, expectation evolution, what works vs. what fails, recurring phrases, and a copyable corrective prompt. - Claude Code fallback: Includes a bundled Python parser that extracts ExitPlanMode denial reasons from ~/.claude/projects JSONL transcripts when no Plannotator archive exists. - Incremental reports and improvement hook: Detects previous reports to analyze only new files, and can write corrective instructions to a hook file injected into future planning sessions. - Use Case: After months of reviewing AI-generated plans, run this Skill to discover that 25% of your denials are for missing narrative overviews, then get a ready-to-paste prompt that fixes it. ## Quick Start Analyze my Plannotator plan archive and generate a compound planning report showing my denial patterns and corrective prompt instructions.
