Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.3.1] - 2026-10-05

plugin/README.md, required by the plugin directory, listing the skills and commands and stating what the plugin reads, writes, and fetches

Commands now register under their documented names (/writing-with-agents:flowers-cycle, :content-strategy, :capture, :writing-setup). The nested commands/ layout had registered them as writing-with-agents:writing:flowers-cycle:COMMAND and similar

Output Frontmatter on the five outer-loop and cross-cutting skills, extending the v0.2.0 provenance chain: knowledge-map (research-intake), domain-whirlybird and content-plan (content-strategist), harvest-report (knowledge-harvester), quality-scorecard (quality-rubric), seo-keyword-map and seo-report (seo-writer). Cross-cutting artifacts attach with applies-to instead of parent
Canonical edit-copy markup convention in skills/_shared/markup-convention.md, shared by Carpenter and Judge
Carpenter handoff templates reference (references/carpenter-handoff.md)
Plugin directory listing metadata in plugin.json: homepage, documentationUrl, supportUrl, privacyPolicyUrl
Privacy policy on the documentation site
Maintainer credit and documentation link at the end of every SKILL.md, with a site footer showing the credit once per page
Representative use cases in docs/examples.md

The plugin now lives in plugin/, and the marketplace source points to ./plugin, so installs ship only skills, commands, references, and LICENSE instead of the whole repository
flowers-cycle command synced with the v0.2.0 skills: optional Fool stress-test after Madman, Architect re-entry from Carpenter, preservation + edit pair delivery and routing at Carpenter, three-signal Judge Consolidated Report with full-rebuild vs light-polish routing
Judge templates and Fool output handling moved to references/judge-consolidated-report.md, replacing the pre-v0.2.0 “Judge Detection Report” template it still held
Architect section mapping format moved into references/blueprint-template.md, which gains STAR/ARROW marks and per-section gaps
Capture note template moved from commands/capture/references/ to references/capture/, linked through ${CLAUDE_PLUGIN_ROOT}
Skill validator accepts 5-7 Core Workflow steps (was exactly 5); the 100-line limit is unchanged and all skills now pass with zero warnings
Skill metadata author corrected and company added; plugin author email corrected
GitHub Actions moved off Node 20: configure-pages v6, upload-pages-artifact v5, deploy-pages v5, action-gh-release v3; the docs site builds on Node 24
GitHub Pages deploy split into its own CI workflow

Install command in README and docs site: writing-with-agents@writing-with-agents (the documented @Jeffallan/writing-with-agents form fails)
Skill validation skips underscore-prefixed directories such as skills/_shared/

YAML frontmatter convention on all inner-loop artifacts (type, version, parent, derived-from) so downstream phases can trace provenance across the Flowers cycle
Preservation + edit copy pair at every Carpenter and Judge delivery: draft-N.md + draft-N-human-edits.md (Carpenter), final-draft-X.md + final-draft-X-human-edits.md (Judge light-polish route)
Carpenter Step 6: post-delivery edit catalog with AskUserQuestion routing the marked-up draft to Architect (structural edits) or Judge (polish edits)
Architect Return-from-Carpenter variant: regenerates the outline as outline-N+1.md when receiving structural rework from the Carpenter
Handling Fool Output sections in Architect (accepts structural revisions) and Judge (accepts tonal-only revisions), with install recommendation pointing to https://github.com/Jeffallan/claude-skills/tree/main/skills/the-fool when the-fool is absent
Madman Step 6: recommended thesis stress-test via the-fool before the Whirlybird phase
Whirlybird Step 1 acknowledges optional Fool pass between Madman and Whirlybird
Judge three-signal aggregation: auto-propagate (strikethroughs and direct rewrites, applied unconditionally), brackets-to-resolve (user commentary with proposed AI resolutions), detection findings (5-pass output)
Judge routing decision (full Carpenter rebuild vs light polish) presented via a single AskUserQuestion call covering every decision at once

Judge consolidated report format now includes auto-propagate and brackets-to-resolve sections alongside detection findings, plus a routing recommendation
Judge Post-Edit Summary template split into two variants (Light Polish route, Full Carpenter Rebuild route)
Carpenter Citation Standard renumbered from Step 6 to Step 7 (now follows the routing step)

10 specialized writing skills across 7 domains (generation, structure, craft, quality, strategy, seo, research)
Inner loop: madman, whirlybird, architect, carpenter, judge, quality-rubric
Cross-cutting: seo-writer
Outer loop: research-intake, content-strategist, knowledge-harvester
31 reference files with deep procedural content
4 workflow commands: flowers-cycle, content-strategy, capture, writing-setup
Validation scripts: validate-skills.py, validate-markdown.py, update-docs.py
Progressive disclosure architecture (Tier 1 SKILL.md + Tier 2 references)
Collaborative oscillation model defining AI/human lead and support roles per phase

Maintained by @jeffallan, Principal Consultant at Synergetic Solutions