Codex Session Kanban

Codex Session Kanban is a static, privacy-aware review surface that turns local Codex CLI / AI-coding sessions into deduplicated task candidates.

The public Pages site is a fixture-only demo. Real .codex logs stay local or in a private repository.

It is not a generic Kanban board and it is not an agent runner. The core job is narrower: help a human quickly review what recent AI sessions were actually doing, which tasks are still active, which older sessions were superseded, and which manual board decisions must not be reverted by later AI sync.

Try it

Open the fixture-only public demo:

https://tenormusica2024.github.io/codex-session-kanban/

In the demo, try this flow:

1. Review the Kanban candidates section.
2. Promote a candidate into the recommended column.
3. Open the card detail panel.
4. Check Why this card exists, evidence categories, extraction timeline, and suppressed predecessor sessions.
5. Move the card status and confirm it becomes a human override lock.
6. Use /, ?, x, j/k, 1-6, c, and b to test keyboard-first triage and handoff copying.

Who this is for

This is useful if you:

• run many Codex / AI-coding sessions and lose track of the real task state
• want an AI secretary to propose a board, but still treat human edits as authoritative
• need to distinguish "needs review", "blocked by user input", "still active", "done", and "dropped"
• care about why a card exists, not only that a transcript summary exists
• want a static/local-first workflow where private logs do not need to leave your machine

License

MIT. See LICENSE.

Release

• Latest release: GitHub Releases
• Changelog: CHANGELOG.md

Contributing / security

• Contributing guide
• Security policy
• Changelog

Privacy stance

This public repository is fixture-only:

• no real .codex session logs
• no local user paths
• no private repository data
• no Vercel bypass tokens

Real session data should stay local or in a private repository. The public demo and screenshots are generated from synthetic fixture data only.

Core differentiation

• Intent-first task extraction: infer the real task from session context instead of using raw prompts like "check progress" or "review this".
• Topic conflict decomposition: separate setup checks, side quests, resolved blockers, and the current main task.
• Parallel tasks inside one repo: keep distinct work streams separate when the repo is the same but the deliverable differs.
• Cross-session task lineage: suppress stale predecessor sessions when a newer session clearly continues or supersedes the same task.
• Human override lock: when a user moves or edits a card, later AI sync should not silently revert that decision.
• Inspectable extraction: detail panels expose evidence categories, extraction timeline, suppressed predecessor sessions, and extraction debug hints.
• Static/private-safe distribution: public demos use fixture data only; real .codex logs stay local/private.
• Provider import normalization: lightweight Claude Code / Cursor / Gemini-style exports can be normalized into the common review schema without adding agent execution scope.
• Fixture coverage diagnosis: public samples are checked for behavior coverage so new fixtures are added only for real extraction/lineage gaps.
• Lightweight prioritization stats: cards expose rough local text-unit estimates, high-activity badges, and large-session hints without becoming a cost dashboard.
• Private scheduled refresh: a separate git-ignored local workflow can scan real .codex sessions without mixing private data into public fixture builds.
• Representative public fixtures: synthetic examples cover topic conflict, same-repo parallel tasks, cross-session predecessor suppression, and non-Codex provider metadata.

Quick start

View the public demo

Open:

https://tenormusica2024.github.io/codex-session-kanban/

Build the fixture locally

npm run local:update

Or call the builder directly:

powershell -ExecutionPolicy Bypass -File .\codex_session_review\build_fixture_snapshot.ps1 -Distribution

Output:

codex_session_review/fixture_snapshot/index.html

Use with your own Codex CLI logs

For a private/local board from your own .codex sessions:

git clone https://github.com/Tenormusica2024/codex-session-kanban.git
cd codex-session-kanban
npm install
npm run private:update:open

This writes the real-session surface under codex_session_review/local_private_review/, which is git-ignored. Do not copy that output into the public fixture build.

Run the release checks

npm install
npm run release:check

To also verify the deployed Pages URL:

npm run release:check:pages

Screenshots

Board overview

Card detail with extraction evidence

Mobile review surface

Mobile detail controls

Docs

• Demo usage guide
• Downloadable artifact usage
• Architecture / data flow
• Import schema
• Provider adapter review
• Fixture coverage diagnosis
• Testing guide
• Local update helper
• Roadmap
• Public release checklist
• Competitive positioning
• Product TODO / adoption policy
• Distribution build notes
• GitHub distribution notes

Smoke tests

Recommended full local check:

powershell -NoProfile -ExecutionPolicy Bypass -File .\codex_session_review\run_public_release_checks.ps1

Or via npm:

npm run release:check

Static artifact check:

python .\codex_session_review\smoke_public_build.py .\codex_session_review\fixture_snapshot\index.html --docs-dir .\codex_session_review\fixture_snapshot\docs --distribution

Downloadable package check:

npm run package:distribution
npm run smoke:package

Browser operation and visible-text check:

npm install
npm run smoke:browser:local
npm run smoke:browser:mobile:local
npm run smoke:browser:narrow:local

Regenerate public README screenshots from fixture data:

npm run local:update
npm run screenshots:update

Lightweight public-fixture refresh helper:

npm run local:update
npm run local:update:open
npm run local:update:full

Private/local real-session refresh helper:

npm run private:update
npm run private:update:open
npm run private:update:full

Provider import diagnosis for observed export shapes:

npm run provider:diagnose

See Private scheduled refresh before wiring this into Windows Task Scheduler.

To verify the public Pages URL:

npm run smoke:browser
npm run smoke:browser:mobile

For the full release check including the deployed Pages URL:

npm run release:check:pages

See Testing guide for details.

GitHub Pages

The workflow .github/workflows/codex-session-kanban-pages.yml builds the fixture with --distribution and can deploy to GitHub Pages.

For public repositories, set:

• Settings → Pages → Source: GitHub Actions

Then push to master or run the workflow manually with deploy_pages=true.