Releases: kenjudy/pdca-agentic-coding-framework
Releases · kenjudy/pdca-agentic-coding-framework
Release list
v1.2.0
Install via Claude UI
- Download
pdca-framework.skillbelow - Go to claude.ai → Customize → Skills → Add Skill
- Upload the file and toggle it on
pdca-framework.skill — TDD-disciplined AI-assisted code generation with Beads integration.
To generate a custom AI workflow skill for any domain, see human-directed-ai-workflow-builder.
See the README for full documentation.
What is Changed
New Skills
- update-changelog: drafts changelog entries from commits ahead of main, groups by category, writes to CHANGELOG.md after confirmation
Repository Split
- human-directed-ai-workflow-builder moved to its own repository
Documentation
- Added CONTRIBUTING.md
- DO phase: added a fifth mandatory Stub check called-shot item
- Renamed skill/Agent.md to skill/SUPERVISION-PROTOCOL.md
- Reordered skill/README.md so Installation precedes Building from Source; release links point to /releases/latest; manual unzip is now the primary path for Claude Code and Codex with matching PowerShell commands
Repository Changes
- Renamed claude-skill/ to skill/, added Codex installation support
- Fixed PowerShell skill descriptor path in build-skill.ps1
- install-skill.ps1 now prompts for install scope
- GitHub repository renamed from kenjudy/pdca-framework to kenjudy/pdca-agentic-coding-framework
Licensing
- Replaced CC0 with a dual license: CC BY 4.0 for docs/prompts, MIT for source code in skill/
Build and Distribution
- Added .claude-plugin/plugin.json manifest for the Stride plugin marketplace
Dependency Updates
- anthropic >=0.116.0, deepeval >=4.1.0, ruff >=0.15.21, mypy >=2.3.0, pytest >=9.1.1, actions/checkout v6 to v7
See CHANGELOG.md for full details and migration notes.
v1.1.0
Install via Claude UI
- Download
pdca-framework.skillbelow - Go to claude.ai → Customize → Skills → Add Skill
- Upload the file and toggle it on
What is Changed
New Skills
- pdca-scaffold: 5-layer Socratic discovery to generate a domain-specific PDCA skill for any complex repeatable human task, with an active learning loop
- daily-retro-pdca: Scaffolded skill for structured daily and weekly retrospection
PDCA Framework Redesigns
- Socratic ACT phase: ACT retrospective replaced with a five-stage Socratic structure
- Claude Code injection system: marker-based injection for Claude Code-specific content (plan mode, think/ultrathink prompts, slash commands)
- Unconditional CHECK and ACT steps: now required items in every plan, not conditional
- Vacuous greens prevention: stub discipline and conditional-first test selection guidance in DO phase
- Slash command prompts added to plan phase prompts
Beads and Workflow
- beads-setup.md: pre-flight check section, MCP server status check, Initializing Beads in a Project section
- beads-workflow.md: Resume a Session section, Export Requirements Document section
- export-requirements.sh: new script to generate a structured requirements document from all open epics and tasks
- Git push is now human-initiated per working agreements; agent only commits
Build and Distribution
- Removed .beads/ from skill distribution
- Renamed src/ to pdca-framework/, flattened core/ directory
- Release workflow: build-scaffold.sh step added, pdca-scaffold.skill attached to GitHub Releases
- Research citations and acknowledgements added to README
See the README for full documentation.
v1.0.2
Install via Claude UI
- Download
pdca-framework.skillbelow - Go to claude.ai → Customize → Skills → Add Skill
- Upload the file and toggle it on
What is Changed
Added
- testing-anti-patterns.md reference file in DO phase, adapted from obra/superpowers (MIT) with PDCA-specific additions covering six common TDD anti-patterns
- SKILL.md now links to the anti-patterns reference from the DO phase description
- CLAUDE.md at repo root incorporating session startup, beads workflow, and supervision rules from AGENTS.md for Claude Code auto-loading
- .gitignore suppresses iCloud sync conflict duplicates
Fixed
- rubric_1b.py clarified that ASCII structural diagrams are not runnable code violations
- Build pipeline (bash + PS1) and test_build.py updated for new anti-patterns file
- Enhanced GitHub Actions metrics workflows with percentile stats and commit quality scoring
See the README for full documentation.
v1.0.1
Internal eval-harness and CI reliability improvements; no user-facing prompt changes.
Install via Claude UI
- Download
pdca-framework.skillbelow - Go to claude.ai → Customize → Skills → Add Skill
- Upload the file and toggle it on
Commits in this release
- Merge pull request #15 from kenjudy/14-refine-eval-process-to-establish-a-reasonable-baseline
- Establish eval baseline: all 15 scenarios passing with majority-vote retry
- Remove Status: must_contain from 3-todos-remaining scenario
- Refine STOP CONDITION: list search targets, do not fabricate findings
- Add STOP CONDITION guard against hallucinated codebase searches
- Constrain 1a output: no library recommendations without codebase evidence
- Fix rubric_1a: score vague-input clarifying questions as 1.0
- Remove called_shot_required from 2-first-step; GEval covers it
- Add skip_geval flag and fix 2-first-step must_not_contain
- Strengthen rubric_2 Process Police exception to be evaluated first
- Simulate post-approval state in Phase 2 eval scenarios
- Fix run-evals.sh to respect test path arguments
- Make run-evals.sh executable
- Skip test_evals.py collection in CI to fix ModuleNotFoundError
- Fix Phase 2 eval scenarios with missing session context
- Upgrade astral-sh/setup-uv to v7 for Node.js 24 compatibility
- Upgrade astral-sh/setup-uv to v6 for Node.js 24 compatibility
- Remove paraphraseable must_contain checks from eval scenarios
- Fix syntax error in run eval shell script
- Add rich file-based output to eval harness
- Update BUILD.md release instructions to reflect automated workflow
- Add release Action and direct download link in README
See the README for full documentation.
v1.0.0 — Initial marketplace release
First release of the PDCA Framework skill for Claude.ai.
Install via Claude UI
- Download
pdca-framework.skillbelow - Go to claude.ai → Customize → Skills → Add Skill
- Upload the file and toggle it on
What's included
- All 5 PDCA phase prompts (Plan 1a/1b, Do, Check, Act)
- Optional Beads integration for cross-session task tracking
- Working agreements reference
See the README for full documentation.