flow
CLAUDE.md0 starsChanged 3 months agoApache-2.0View on GitHub
A Claude Code plugin: a phased development workflow (setup → research → plan → gate 1 → implement → review → gate 2 → wind-down) whose gates are enforced by a PreToolUse hook rather than by prompt discipline. See README.md for usage and docs/design.md for why it's built that way. Layout: skills/ (phase skills), agents/ (subagents), hooks/ (the validator and its shell-command reader flow_shell.py, the snapshot script, the worktree opener, and the display mod in hooks/mod/), types/ (the mod's state…
In this repository9 other files it gives its agents.
- flow-implementSkill
- flow-planSkill
- flow-reportSkill
- flow-researchSkill
- flow-reviewSkill
- flow-shapeSkill
- flowSkill
- flow-startSkill
- flow-wrapSkill
The file
We split this CLAUDE.md into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
Every PR bumps the plugin version.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
flow — repo guidance
A Claude Code plugin: a phased development workflow (setup → research → plan →
gate 1 → implement → review → gate 2 → wind-down) whose gates are enforced by a
PreToolUse hook rather than by prompt discipline. See README.md for usage and
docs/design.md for why it's built that way.
Layout: skills/ (phase skills), agents/ (subagents), hooks/ (the
validator and its shell-command reader flow_shell.py, the snapshot script,
the worktree opener, and the display mod in hooks/mod/), types/ (the mod's state contract), templates/ (files the
skills write into target repos), .claude-plugin/ (manifests), scripts/
(scripts/demo/ retakes the README's screenshots), tests/.
Version bumps — every change, no exceptions
Every PR bumps the plugin version. Installed copies have no other signal that they are stale, so an unbumped change ships invisibly.
Bump all three version fields together — they must always be equal:
.claude-plugin/plugin.json→version.claude-plugin/marketplace.json→metadata.version.claude-plugin/marketplace.json→plugins[0].version
Which digit:
- patch (0.3.0 → 0.3.1) — fixes, docs, test-only changes, wording that doesn't change what a skill does.
- minor (0.3.0 → 0.4.0) — a new skill, agent, or template; any change to what the workflow asks of the user or what the validator blocks.
- major — a breaking change to state-file format, config, or gate semantics: an in-flight task in a target repo would break on upgrade.
Bumping is the last edit before committing, so it reflects everything in the PR. One bump per PR, not one per commit.
Checks
python3 -m unittest discover tests -v # full suite
claude plugin test . # the mod (tests/mod/)
claude plugin validate . # manifest + mod scan
python3 scripts/check_version_bump.py # manifests agree
python3 scripts/check_version_bump.py main # ...and are ahead of mainCI (.github/workflows/version-bump.yml) runs the same script on every PR and
fails it when the version is unchanged or the manifests disagree. Escape hatch,
user-only: label the PR no-version-bump. The check must be marked required in
branch protection for main to actually block a merge.
README images
The images in docs/images/ are captured from a live Claude Code session,
never drawn by hand. Retake them in the PR that changes what they show:
| Image | Shot | Shows | Retake after changing |
hero.png | hero | the rail mid-implementation | the band (hooks/mod/band.tsx, look.tsx), the snapshot (hooks/flow_status.py) |
gate.png | gate | the rail at gate 1 | the band, gate actions |
gate.gif | gif | approving gate 1 (one model turn) | the band, gate actions, phase toasts |
blocked.png | blocked | a refused edit (one model turn) | the validator's message, the refusal toast |
pane-progress.png, pane-tasks.png | pane-progress, pane-tasks | /flow-pane | hooks/mod/pane.tsx |
status.png | status | /flow-status | hooks/mod/status.tsx, hooks/flow_status.py |
split.png | split | t opening a terminal | hooks/flow_open.py, the band's key hints |
setup.png | setup | /flow-term setup (iTerm only) | hooks/flow_iterm.py, the setup checklist |
scripts/demo/shoot.py --check # what's missing, with install commands
scripts/demo/shoot.py # every shot, into docs/images (~6 min)
scripts/demo/shoot.py gate hero # just these- It loads flow from this checkout, so run it from the branch or worktree being changed, after the code change.
- It needs tmux, freeze, Chrome or Chromium, claude and ffmpeg. The font is fetched and hash-checked on first run.
- The
setupshot runs only inside iTerm with its Python API on. Elsewhere it is skipped and the old image stays. - Seven shots are byte-identical from run to run, so if one of their files
changed, the picture changed.
blockedandgifinclude a model turn and differ every run, so retake them only when what they show has changed. - Open every image it wrote before committing. Each one should show what the README text next to it says, with no stray toasts, a light theme or cut-off rows. If the README text no longer matches the UI, fix the text too.
- The demo repo is rebuilt from
scripts/demo/make-demo-repo.sheach run. Change the fictional tasks there, not by hand.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

