feat(skill): compound planning Claude Code fallback + blog post (#461)

* feat(skill): add Claude Code fallback to compound planning skill

Plannotator remains the first-class data source. When users don't have a
Plannotator archive, the skill now falls back to extracting ExitPlanMode
denial reasons from Claude Code JSONL transcripts via a bundled parser.

Adds Source Selection section, per-phase fallback notes, a progressive
disclosure reference (references/claude-code-fallback.md), and the
extraction script (scripts/extract_exit_plan_mode_outcomes.py).

For provenance purposes, this commit was AI assisted.

* fix(skill): remove duplicate detection logic and fix fallback skip

Phase 0 was duplicating the Source Selection branching and telling
fallback mode to skip Previous Report Detection. Now Phase 0 defers
to the already-chosen mode and both paths proceed to report detection.

For provenance purposes, this commit was AI assisted.

* blog: add Continuously Improve Claude Code Plans post

Introduces compound planning to users — how denial feedback compounds
into better future plans, works for both Plannotator and plain Claude
Code users. Video placeholder included for later replacement.

For provenance purposes, this commit was AI assisted.

* blog: revise compound planning post and adjust prose typography

Rework blog copy, add looping video embeds from CloudFront media
bucket, simplify process overview, add install link. Bump prose
font-size to 0.95rem and line-height to 1.6 on desktop.

For provenance purposes, this commit was AI assisted.

* blog: move intro video, copy tweaks, video spacing

For provenance purposes, this commit was AI assisted.

Author: backnotprop

apps/marketing/src/content/blog/continuously-improve-claude-code-plans.md

@@ -0,0 +1,55 @@
+---
+title: "Continuously Improve Claude Code Plans"
+description: "Use your denial history to find feedback patterns, generate a personalized report, and automatically improve every future plan."
+date: 2026-04-01
+author: "backnotprop"
+tags: ["compound-planning", "plan-mode", "claude-code"]
+---
+
+**Compound Planning.** If you've been actively using plan mode in Claude Code, there's an opportunity to improve how your agents plan for you. Here is a skill that analyzes your plan denial history, surfaces your own feedback patterns, and creates an automated loop that refines planning over time (at your request). The point is to consistently optimize what works best for you.
+
+<video width="100%" style="aspect-ratio: 16/9; border-radius: 8px; margin-bottom: 1.5rem;" autoplay loop muted playsinline controls>
+  <source src="https://d17ygohy796f9l.cloudfront.net/videos/compound-planning.mp4" type="video/mp4" />
+</video>
+
+If you use plan mode and deny plans, you already have the data. The skill reads it and puts it to work. It works day one for any Claude Code user who has been actively using plan mode. [Plannotator](https://github.com/backnotprop/plannotator) users get an even richer analysis since Plannotator captures full plan text and inline annotations with every denial. Plannotator is [open source](https://github.com/backnotprop/plannotator) and free.
+
+[Install Plannotator](https://github.com/backnotprop/plannotator?tab=readme-ov-file#install-for-claude-code), then run:
+
+```
+/plannotator-compound
+```
+
+## How it works
+
+When you deny a plan in Claude Code, that interaction gets logged. If you're a Plannotator user, it's even better. The more you use it, the richer your archive gets. Full plan text, inline annotations, structural feedback, all saved as markdown files that build up over time.
+
+Compound Planning reads that archive in three steps:
+
+1. **Analyze** all your plan denials and approvals, surfacing insight patterns
+2. **Report** generates a personalized report so you can view your own findings and insights
+3. **Hook** creates an improvement hook that gets injected on plan enter mode, carrying forward the insights from the last time you ran the skill
+
+The report is personalized. Every quote is pulled from your actual feedback. Every percentage is calculated from your real data. The corrective instructions at the end trace directly back to your most frequent denial reasons.
+
+<video width="100%" style="aspect-ratio: 16/9; border-radius: 8px; margin-bottom: 1.5rem;" autoplay loop muted playsinline controls>
+  <source src="https://d17ygohy796f9l.cloudfront.net/videos/compound-planning-report.mp4" type="video/mp4" />
+</video>
+
+## Works for all Claude Code users
+
+Plannotator users get the richest analysis since the archive contains full plan text and inline annotations. But you don't need Plannotator.
+
+If you use Claude Code with plan mode, your denial reasons live in `~/.claude/projects/`. The skill includes a Python parser that extracts your `ExitPlanMode` outcomes, filters out boilerplate, and produces clean records of your human-authored feedback. The same analysis pipeline runs on this data and produces the same report.
+
+## The feedback loop
+
+The real value is the improvement hook. The corrective instructions from your report can be saved to a file that gets injected into every future planning session automatically. Claude sees your feedback patterns before writing any plan.
+
+Your denied plans aren't wasted work. They're the specification for better plans going forward.
+
+## Try it
+
+If you have [Plannotator](https://github.com/backnotprop/plannotator) installed, it works out of the box. If you're using Claude Code without Plannotator, the skill works with your existing session logs, no additional setup.
+
+The more denial history you have, the richer the analysis. Start with the [installation guide](/docs/getting-started/installation/) or check out the [repo](https://github.com/backnotprop/plannotator).

apps/marketing/src/styles/global.css

@@ -37,6 +37,13 @@
   color: var(--foreground);
 }
 
+@media (min-width: 768px) {
+  .prose {
+    font-size: 0.95rem;
+    line-height: 1.6;
+  }
+}
+
 .prose h1 {
   font-size: 1.5rem;
   font-weight: 700;

apps/skills/plannotator-compound/SKILL.md

@@ -4,7 +4,8 @@ disable-model-invocation: true
 description: >
   Analyze a user's Plannotator plan archive to extract denial patterns, feedback
   taxonomy, evolution over time, and actionable prompt improvements — then produce
-  a polished HTML dashboard report.
+  a polished HTML dashboard report. Falls back to Claude Code ExitPlanMode denial
+  reasons when Plannotator data is unavailable.
 ---
 
 # Compound Planning Analysis
@@ -16,15 +17,37 @@ them into actionable insights, and produce an elegant HTML dashboard report.
 This is a multi-phase process. Each phase must complete fully before the next begins.
 Research integrity is paramount — every file must be read, no skipping.
 
+## Source Selection
+
+Before starting the analysis, determine which data source is available.
+
+1. **Plannotator mode (first-class)** — Check `~/.plannotator/plans/`. If it
+   exists and contains `*-denied.md` files, use this mode. The entire workflow
+   below is written for Plannotator data.
+
+2. **Claude Code fallback mode** — If the Plannotator archive is absent or
+   contains no denied plans, check `~/.claude/projects/`. If present, read
+   [references/claude-code-fallback.md](references/claude-code-fallback.md)
+   before continuing. That reference explains how to use the bundled parser at
+   [scripts/extract_exit_plan_mode_outcomes.py](scripts/extract_exit_plan_mode_outcomes.py)
+   to extract denial reasons from Claude Code JSONL transcripts. Every phase
+   below has a short note explaining what changes in fallback mode — the
+   reference file has the details.
+
+3. **Neither available** — Ask the user for their Plannotator plans directory or
+   Claude Code projects directory. Do not guess.
+
 ## Phase 0: Locate Plans & Check for Previous Reports
 
-Check for `~/.plannotator/plans/`. If it exists and contains `.md` files, proceed.
+Use the mode chosen in Source Selection above.
 
-If not found, ask the user:
-> "I couldn't find plans at ~/.plannotator/plans/. Where is your plans directory?"
+**Plannotator mode:** Verify the plans directory contains `*-denied.md` files. If
+none exist, fall back to Claude Code mode before stopping.
 
-Once you have the path, verify it contains files matching `*-denied.md`. If none
-exist, inform the user there's no denial data to analyze and stop.
+**Claude Code fallback mode:** Run the bundled parser per the fallback reference to
+build the denial-reason dataset. Create `/tmp/compound-planning/` if needed.
+
+In either mode, proceed to Previous Report Detection below.
 
 ### Previous Report Detection
 
@@ -112,6 +135,11 @@ Extract dates from all filenames.
 
 Tell the user what you found and that you're beginning the extraction.
 
+**Claude Code fallback mode:** The Plannotator inventory fields above do not apply.
+Follow the inventory instructions in
+[references/claude-code-fallback.md](references/claude-code-fallback.md) instead —
+report the denial-reason dataset assembled by the parser.
+
 ## Phase 2: Map — Parallel Extraction
 
 This is the most time-intensive phase. You must read EVERY `*-denied.md` file
@@ -121,6 +149,11 @@ This is the most time-intensive phase. You must read EVERY `*-denied.md` file
 files dated after the cutoff date if running incrementally. In incremental mode,
 only process files whose embedded YYYY-MM-DD date is strictly after the cutoff.
 
+**Claude Code fallback mode:** The parser output is the clean source dataset. Read
+the fallback reference for the extraction prompt and batching strategy specific to
+JSON part files. Do not go back to raw `.jsonl` logs unless the parser fails or the
+user asks for audit-level verification.
+
 **Important:** Only read `*-denied.md` files. Do NOT read approved plans,
 annotation files, or diff files. Each denied file contains the full plan text
 followed by a `---` separator and the reviewer's feedback — everything needed
@@ -248,6 +281,10 @@ extraction files and produce the full analysis. This covers most datasets.
    merges taxonomies, combines counts, deduplicates patterns, and reconciles any
    conflicting categorizations across partials.
 
+**Claude Code fallback mode:** The reduction phase is the same. The only upstream
+difference is that extraction files were derived from normalized denial-reason JSON
+instead of Plannotator markdown files.
+
 ### Reduction Prompt
 
 Give each reduction agent this prompt (adapt file paths for single vs multi-stage):
@@ -427,6 +464,10 @@ one — the flow moves from "what happened" through "why" to "what to do about i
 - If the user has < 3 months of data, reduce the evolution section to fewer cards
 - If most denied files lack feedback below the `---` (bare denials with no
   annotations), note this in the narrative — the analysis will be thinner
+- **Claude Code fallback mode:** Explicitly label the report source as Claude Code
+  `ExitPlanMode` denial reasons. Do not fabricate Plannotator-only fields such as
+  annotation counts or approved-plan line counts. See the fallback reference for
+  KPI substitutes and footer/provenance guidance.
 - If fewer than 5 denial categories emerge, combine the taxonomy and patterns
   sections into one
 - If the dataset is very small (< 20 files), the narrative should acknowledge the
@@ -457,6 +498,10 @@ Tell the user:
 - Where the report was saved (including version number)
 - If incremental: remind the user that earlier findings are in the previous report
 
+**Claude Code fallback mode:** Adapt the summary per the fallback reference —
+report human denial reasons analyzed and total `ExitPlanMode` attempts scanned
+instead of Plannotator file counts.
+
 ## Phase 6: Improvement Hook
 
 After presenting the summary, ask the user if they want to enable an **improvement
@@ -510,6 +555,8 @@ Read the existing file and present the user with a choice:
 
 ## Important Notes
 
+- **Data source priority:** Plannotator is the first-class path. Claude Code log
+  analysis is the secondary path for users without Plannotator archives.
 - **Research integrity:** Every file must be read. The value of this analysis comes
   from completeness. Sampling or skipping undermines the findings.
 - **Real data only:** Never fabricate quotes, percentages, or patterns. If the data