Skip to content

ogfrench/agentic-workflow

v2026.11.11

Source-grounded software engineering workflows for architecture, implementation review, research and release delivery. Uses only tools available in the current host.

Agentic Workflow for ChatGPT

Full portable edition, v2026.11.11. Source: ogfrench/agentic-coding-setup.

This package provides the 23 source-owned engineering, story and design skills from the Claude Code edition of agentic-workflow, 4 previously portable review/setup skills, and the curated public engineering-practice knowledge base. All skills retain their owning source guidance and supporting references. A short ChatGPT host-adapter contract is injected into the Claude-derived skills to prevent claiming unavailable tools or runtime features.

Included workflow domains

  • Architecture Gardener, Delivery Loop, Issue Management and Product Passport.
  • Best Practices, Learning Loop, Knowledge Base Builder and Model Delegation.
  • Visual Loop, Session Continuity, Token Efficiency, GitHub Actions Cost and Netlify Cost.
  • Test Strategy and Security Review. Before/after proof and production readiness are references in existing skills.
  • Plugin Experience Design, Presentation Story, Presentation Design and Design System Adapter. Visual Loop also includes visual restraint.
  • Frogify, Deslopify and Humanizer for selective brand writing and editorial revision, with Humanizer's MIT notice.
  • Adversarial Review, Implementation Proof, Release Handover and Setup Audit from the former Lite edition.

Select a skill for a relevant task. Ordinary chat should not activate all skills. Read supporting references only if relevant and accessible.

The standalone frog-writing and presentation-craft packages are retired; the writing and presentation skills are included in both coordinated editions. Use frogViz's actual authoring workflow whenever explicitly requested, rather than a generic imitation. Existing legacy plugin installations are not silently removed.

What this does not install

No Claude Code hooks, commands, subagents, user setup scripts, local MCP processes, credentials, automatic project sync, or autonomous background work. ChatGPT tools, local file access, authentication, runtime and UI/browser capabilities are independent. A skill's instructions never create tool permissions.

Knowledge consists of source-linked, curated engineering practices and their public index, not client projects or a private file corpus. The source repository's code and runtime behavior remain authoritative; documentation may be stale.

Version and deprecation

agentic-workflow v2026.11.11 supersedes ChatGPT Workflow Lite v1.12.0 and the legacy personal ogfrench-workflow v0.1.1 for ChatGPT. The Lite archives remain only as historical compatibility packages and are deprecated, not auto-deleted. The Claude Code edition of agentic-workflow (named claude-workflow through 2026.11.3) shares this release number, but remains independently packaged and host-verified.

Building and testing

Run python3 -m unittest discover -s test -p test_agentic_workflow_chatgpt.py, then python3 scripts/package_agentic_workflow_chatgpt.py. The build uses the checked-in canonical source, fails on missing skills or linked support files, keeps an integrity manifest and creates a reproducible ZIP. From the next GitHub release after v2026.11.9, obtain the versioned ChatGPT ZIP and shared SHA256SUMS.txt from GitHub Releases. Verify the digest, then create or update the same-identity plugin privately through ChatGPT Plugin Creator. Inspect installed version, all 27 skills and a real repository task. The public archive does not publish the ChatGPT account plugin. Claude Code installation is separate: use bash setup-user.sh from the reviewed repository checkout and inspect claude plugin list.

No public directory submission or share permission is implied.