Claude/optimistic dubinsky

.claude-plugin/marketplace.json

@@ -0,0 +1,36 @@
+{
+  "name": "pagekit",
+  "owner": {
+    "name": "Hiten Shah"
+  },
+  "metadata": {
+    "description": "PageKit: skills for making web pages that are contextually relevant to your audience and what your product does.",
+    "version": "0.3.0"
+  },
+  "plugins": [
+    {
+      "name": "pagekit",
+      "source": {
+        "source": "github",
+        "repo": "hnshah/pagekit"
+      },
+      "description": "Skills, subagents, and scripts for the PageKit method: signal doc, message spine, first-page decision, page argument shape, proof map, first-page draft, claim check.",
+      "version": "0.3.0",
+      "author": {
+        "name": "Hiten Shah"
+      },
+      "homepage": "https://github.com/hnshah/pagekit",
+      "repository": "https://github.com/hnshah/pagekit",
+      "license": "MIT",
+      "keywords": [
+        "pagekit",
+        "content-strategy",
+        "copywriting",
+        "landing-pages",
+        "claim-check",
+        "anti-slop"
+      ],
+      "category": "productivity"
+    }
+  ]
+}

.claude-plugin/plugin.json

@@ -0,0 +1,19 @@
+{
+  "name": "pagekit",
+  "description": "Skills for making web pages that are contextually relevant to your audience and what your product does. Ships the 7-step method (signal doc, message spine, first-page decision, page argument shape, proof map, first-page draft, claim check) plus tooling skills and subagents.",
+  "version": "0.3.0",
+  "author": {
+    "name": "Hiten Shah"
+  },
+  "homepage": "https://github.com/hnshah/pagekit",
+  "repository": "https://github.com/hnshah/pagekit",
+  "license": "MIT",
+  "keywords": [
+    "pagekit",
+    "content-strategy",
+    "copywriting",
+    "landing-pages",
+    "claim-check",
+    "anti-slop"
+  ]
+}

.github/PULL_REQUEST_TEMPLATE.md

@@ -16,7 +16,7 @@ Delete the comments before submitting.
 - [ ] `bash scripts/doctor.sh` → PASS
 - [ ] `bash scripts/slop-check.sh` → exit 0 clean
 - [ ] (if the PR touches run structure) `bash scripts/run-check.sh runs/<name>` → FULLY LOGGED (or tier you target)
-- [ ] (if the PR touches prompts/frameworks) reviewed against `AGENTS.md` and `CLAUDE.md` for consistency
+- [ ] (if the PR touches skills or their bundled references) reviewed against `AGENTS.md` and `CLAUDE.md` for consistency
 
 ## Notes for the reviewer
 

AGENTS.md

@@ -1,94 +1,55 @@
 # AGENTS.md
 
-Neutral agent contract for PageKit. Readable by Codex (primary `AGENTS.md` consumer), Claude Code (which also reads `CLAUDE.md`), Claude Cowork, and any agent reading the repo cold.
+Neutral agent contract for PageKit. Readable by Codex, Claude Code, and any agent reading the repo cold.
 
 ## What PageKit is
 
-A system for producing better pages by strengthening the source materials that shape the copy. Markdown-first. Method-over-prompts.
+A Claude Code plugin plus a chat-paste path for producing web pages that are contextually relevant to your audience and what your product does. The method runs upstream of copy: signal → spine → first-page decision → argument shape → proof map → first-page draft → claim check.
 
-## Where to start
-
-Read these files in this order:
+## If you are Claude Code
 
-1. **`pagekit.yaml`** — canonical method manifest. The source of truth for every step, artifact, framework, template, guided-run and prompt. If you need to know what step N does and where its output goes, this file tells you.
-2. **`CLAUDE.md`** — fuller operational contract, including the quality rules, the anti-slop rules, and the "Running PageKit end-to-end as an agent" section.
-3. **`README.md`** — human-facing overview.
+The plugin is self-contained in this repo. Skills live under [`skills/`](skills/), subagents under [`agents/`](agents/), and the plugin manifests under [`.claude-plugin/`](.claude-plugin/). Invoke the orchestrator with `Run PageKit on <object>` and it dispatches the 7-step chain.
 
-## Method in one glance
+Key entry points:
+- `skills/pagekit/SKILL.md`: orchestrator
+- `skills/pagekit-<step>/SKILL.md`: per-step skills (signal-doc through claim-check)
+- `agents/pagekit-claim-checker.md`: adversarial claim-check subagent
+- `agents/pagekit-evaluator-pass.md`: adversarial evaluator-pass subagent
 
-```
-signal-doc → message-spine → first-page-decision → page-argument-shape → proof-map → first-page-draft [ → claim-check ]
-```
+## If you are Codex or another chat-based agent
 
-Canonical prompts: `prompts/01-signal-doc.md` through `prompts/07-claim-check.md`.
+Use the prompt-paste path. Open [`prompt-paste/`](prompt-paste/) and walk the 7 prompts in order. Each step consumes the previous step's output.
 
-## Running a fully-logged run
+## Running a fully-logged run from the shell
 
 ```sh
-scripts/new-run.sh <run-name>           # scaffold the folder
-# … work through the steps, filling prompts/NN-*.md and outputs/NN-*.md
-scripts/run-check.sh runs/<run-name>    # validate the run meets the fully-logged tier
-```
+# scaffold the folder
+bash scripts/new-run.sh <run-name>
 
-Or via make:
+# …work through prompts/NN-*.md and outputs/NN-*.md
 
-```sh
-make new-run NAME=<run-name>
-make run-check RUN=<run-name>
-```
-
-## Anti-slop
-
-Drafts must respect `frameworks/anti-slop.md`. The regression check is `scripts/slop-check.sh`. Run it on the draft before the claim-check step; run it again after.
-
-## Claim-check
-
-```sh
-scripts/claim-check.sh runs/<run>/first-page-draft.md runs/<run>/proof-map.md --severity hard
-```
+# validate against the logging tiers
+bash scripts/run-check.sh runs/<run>
 
-Outputs a ready-to-paste prompt. Paste into the model, then save the result as `runs/<run>/claim-check.md` and the corrected draft as `runs/<run>/first-page-draft-corrected.md`.
+# claim-check the draft at the severity you intend
+bash scripts/claim-check.sh runs/<run>/first-page-draft.md runs/<run>/proof-map.md --severity hard
 
-## Pre-flight
+# heuristic anti-slop regression against the draft
+bash scripts/slop-check.sh runs/<run>/first-page-draft.md
 
-```sh
-scripts/doctor.sh
-# or
-make doctor
+# pre-flight check for the repo
+bash scripts/doctor.sh
 ```
 
-## If you are Claude Code
-
-See `.claude/skills/pagekit/SKILL.md` for the orchestrator skill. Per-step skills live under `.claude/skills/pagekit-<step>/`. The custom subagent for claim-check is `.claude/agents/pagekit-claim-checker.md`. The SessionStart hook in `.claude/settings.json` runs `scripts/doctor.sh` on session start.
-
-Agentic path: `agentic/claude-code-agentic-path.md`.
-
-## If you are Codex
-
-This file is your contract. Read `pagekit.yaml` for the method. Scripts in `scripts/` are the tooling you will drive. Custom prompts / skills for Codex live alongside your own installation (see Codex docs); the canonical prompts in `prompts/` are what they should wrap.
-
-Agentic path: `agentic/codex-agentic-path.md`.
-
-## If you are Claude Cowork
-
-The skill bundle at `.claude/skills/` is the plugin body. See `agentic/claude-cowork-agentic-path.md` for local installation and eventual published-plugin notes.
-
 ## Hard rules
 
-Non-negotiable across tools and agents:
-
-- **Do not hard-code homepage-first.** Step 03 is where the first page is chosen. The object decides.
+- **Do not hard-code homepage-first.** Step 03 picks the first page from the object. Name rejected candidates.
 - **Do not invent proof.** Mark unverified product-specific claims with `*[verification flag: ...]*` inline.
-- **Do not ship a draft with AI-slop patterns.** See `frameworks/anti-slop.md`. Run `scripts/slop-check.sh` and resolve every hit.
-- **Do not skip run logging.** Fully-logged runs are defined in `frameworks/run-logging.md`. If you are producing a run that claims to be fully-logged, it must meet that tier. `scripts/run-check.sh` tells you if it does.
-- **Edit the canonical prompts in `prompts/`, never duplicate.** Guides reference them.
+- **Do not ship a draft with AI-slop patterns.** See `skills/pagekit/references/anti-slop.md`. Run `scripts/slop-check.sh` and resolve every hit.
+- **Do not skip run logging.** Fully-logged runs are defined in `skills/pagekit/references/run-logging.md`. `scripts/run-check.sh` tells you if your run meets the tier.
 
 ## Getting unstuck
 
 - Output of a step feels generic → fix upstream source material, not the step output.
 - First-page decision reluctant → add a mechanism / proof / comparison brief in `sources/` and rerun.
-- Claim-check flagging the same pattern every run → update `frameworks/anti-slop.md` and the argument-shape drafting constraints so the pattern does not return.
-
-## What this file is not
-
-A full user manual. The full method lives in `frameworks/` and `guided-runs/`. The tool-specific paths live in `guides/`, `quickstart/` and `agentic/`. This file is the doormat that tells you where to look.
+- Claim-check flagging the same pattern every run → update `skills/pagekit/references/anti-slop.md` and the argument-shape drafting constraints so the pattern does not return.

CHANGELOG.md

@@ -6,6 +6,40 @@ Format loosely follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## [Unreleased]
 
+## [0.3.0] — 2026-04-17
+
+Skills-first restructure. PageKit now ships as a Claude Code plugin: one install, twelve skills, two subagents, a chat-paste path for non-Claude-Code chat windows. Breaking for the repo layout; non-breaking for the method itself.
+
+### Added
+- `.claude-plugin/plugin.json` and `.claude-plugin/marketplace.json` — plugin manifests. Install via `/plugin marketplace add hnshah/pagekit` + `/plugin install pagekit@pagekit`.
+- `skills/` top-level directory with 12 skills (orchestrator, 7 method steps, 4 tooling skills). Each skill has its own bundled `references/` folder with framework, template, and prompt.
+- `skills/pagekit/references/` owns the shared references (`anti-slop.md`, `run-logging.md`, `method.md`). Sibling skills link by relative path.
+- `agents/` — subagents live at repo root instead of under `.claude/agents/`.
+- `examples/` — worked runs moved here from `runs/` (both `vegan-dog-food-verdel` and `personal-crm-founders`). `runs/` stays as user scratch.
+- `prompt-paste/` — seven prompts stripped of YAML frontmatter for pasting into ChatGPT / Perplexity / Grok / Gemini / other chat windows.
+
+### Changed
+- `README.md` rewritten skills-first: install, skills list, worked example. One page, no tree of entry choices.
+- `AGENTS.md` and `CLAUDE.md` slimmed to neutral agent contracts (~45 lines each). Point at skills and the plugin install.
+- `CONTRIBUTING.md` rewritten against the new layout.
+- `scripts/new-run.sh` resolves canonical prompts from `skills/pagekit-<step>/references/prompt.md`.
+- `scripts/claim-check.sh` reads from `skills/pagekit-claim-check/references/prompt.md`.
+- `scripts/slop-check.sh` scans `runs/` and `examples/` by default.
+- `scripts/doctor.sh` validates the plugin layout: plugin manifest, top-level `skills/` and `agents/`, bundled references.
+- `scripts/run-check.sh` — fix an unbound-variable warning at PUBLISHABLE tier (cosmetic; tier output was always correct).
+
+### Removed
+- `START-HERE.md`, `quickstart/`, `guides/`, `guided-runs/`, `agentic/` — replaced by the rewritten README and the `prompt-paste/` path.
+- `frameworks/`, `templates/`, `prompts/` — absorbed into each skill's `references/` folder.
+- `pagekit.yaml` — distilled into `skills/pagekit/references/method.md`.
+- `.claude/skills/` and `.claude/agents/` — migrated to the top-level `skills/` and `agents/`.
+- `templates/output-judgment-template.md` — orphan, not referenced by any step.
+
+### Method integrity
+The anti-slop framework, run-logging tiers, claim-check severities, and all seven method steps are unchanged from v0.2. Only the filesystem layout and the install surface changed. `scripts/slop-check.sh` and `scripts/run-check.sh` on both example runs still pass at their prior tiers.
+
+## [0.2.0] — 2026-04-15
+
 ### Added
 - `.github/workflows/check.yml` runs `doctor.sh` and `slop-check.sh` on every push and PR.
 - `.github/PULL_REQUEST_TEMPLATE.md` + issue templates (`method-feedback`, `bug`, `method-proposal`).
@@ -46,5 +80,7 @@ Baseline public release. The agentic foundation is in place.
 ### Notes
 - PRs #1–#8 all merged prior to this release. See the GitHub PR history for the run-to-repo-improvement loop that produced the method surface.
 
-[Unreleased]: https://github.com/hnshah/pagekit/compare/v0.1.0...HEAD
+[Unreleased]: https://github.com/hnshah/pagekit/compare/v0.3.0...HEAD
+[0.3.0]: https://github.com/hnshah/pagekit/compare/v0.2.0...v0.3.0
+[0.2.0]: https://github.com/hnshah/pagekit/compare/v0.1.0...v0.2.0
 [0.1.0]: https://github.com/hnshah/pagekit/releases/tag/v0.1.0