Init
Bootstrap a new project with the Flow workflow - creates AGENTS.md from the template, creates .claude/tracker.json from the example, and optionally sets up cross-vendor review profiles. Run once per project after installing the Flow plugin. Explicit-only; do not auto-fire.From its SKILL.md
npx -y skills add corticalstack/flow --skill initAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 1 command, including `bash ${CLAUDE_PROJECT_DIR}/.claude/scripts/tracker.sh setup-labels`.
SKILL.md
5.2 KB, ~1.3k tokens by cl100k_base, as published. Nobody here has run it
Flow init
You are bootstrapping the current user project (the directory at ${CLAUDE_PROJECT_DIR}) with Flow's workflow conventions and config templates. Read templates from the plugin install at ${CLAUDE_PLUGIN_ROOT}. Treat this as an interactive setup wizard: ask before each write, never overwrite an existing file.
Steps
1. AGENTS.md
Check if ${CLAUDE_PROJECT_DIR}/AGENTS.md exists.
-
If it does NOT exist:
- Read
${CLAUDE_PLUGIN_ROOT}/docs/AGENTS.md.template. - Ask the user: "I'll create AGENTS.md at your project root from Flow's template. It has a
[Replace this section ...]placeholder for the project overview - I can fill it in if you give me a one-paragraph description of this project, or leave the placeholder for you to edit later. What would you like?" - Resolve the placeholder if the user gives a description, then write the content to
${CLAUDE_PROJECT_DIR}/AGENTS.md.
- Read
-
If it DOES exist:
- Read it. Tell the user: "AGENTS.md already exists. Flow expects (1) the feature-branch requirement, (2) the research / plan / implement / validate stages with
flow/artifacts, (3) the workflow-state vocabulary, (4) a pointer to the tracker adapter. The template is at${CLAUDE_PLUGIN_ROOT}/docs/AGENTS.md.template- compare and merge anything missing." - Do NOT overwrite.
- Read it. Tell the user: "AGENTS.md already exists. Flow expects (1) the feature-branch requirement, (2) the research / plan / implement / validate stages with
2. Tracker config
Check if ${CLAUDE_PROJECT_DIR}/.claude/tracker.json exists.
-
If it does NOT exist:
- Ask the user: "Which issue tracker does this project use?
github,azure-devops, ornone? I'll create.claude/tracker.jsonfrom the Flow example with your choice as the active backend." - Read
${CLAUDE_PLUGIN_ROOT}/.claude/tracker.example.json. - Build the project's
.claude/tracker.jsonwith"tracker": "<user-choice>"and the corresponding backend block populated as much as you can from the example (keep the other backend blocks as inert reference - the adapter only reads the active one). - Create the
${CLAUDE_PROJECT_DIR}/.claude/directory if it does not exist, then write the file. - Then per backend:
github: tell the user to confirmgh auth loginis done, then offer the label step below.azure-devops: warn that the backend is currently a stub (PR 2 pending) - tag / state-transition writes will not yet happen.none: tell the user nothing more is needed; mutating tracker calls become no-ops, queries return empty JSON.
- Ask the user: "Which issue tracker does this project use?
-
If it DOES exist:
- Tell the user: ".claude/tracker.json already exists; leaving it alone. If you want to change backend, edit it directly - see
${CLAUDE_PLUGIN_ROOT}/.claude/tracker.example.jsonfor the schema." - Read it to determine the active backend - the label step below still applies if it is
github.
- Tell the user: ".claude/tracker.json already exists; leaving it alone. If you want to change backend, edit it directly - see
2a. Workflow-state labels (github backend only)
Only run this when the active backend is github. Skip it entirely for azure-devops (state tags are free-form, created on first transition) and none.
- Ask the user: "Create the 8 workflow-state labels in this repo now? This writes to the GitHub repo, so
gh auth loginmust already be done. (yes / no)" - If yes, run
bash ${CLAUDE_PROJECT_DIR}/.claude/scripts/tracker.sh setup-labels. The command is idempotent (gh label create --force), so re-running is safe. If it fails (e.g.ghnot authenticated), print the exact error and tell the user to rungh auth loginand re-run the command - do not run the auth flow yourself. - If no, tell the user they can create them later with
bash .claude/scripts/tracker.sh setup-labels.
3. Cross-vendor review (optional)
Ask: "Want to set up cross-vendor review now? It's optional - you can do it later. (yes / no)"
- If yes:
- Check if
${CLAUDE_PROJECT_DIR}/.claude/skills/cross-review/profiles.jsonexists. - If not, read
${CLAUDE_PLUGIN_ROOT}/.claude/skills/cross-review/profiles.example.jsonand write a copy to${CLAUDE_PROJECT_DIR}/.claude/skills/cross-review/profiles.json(creating the directory if needed). - Tell the user to edit the file with their Foundry endpoint + deployment, GitHub Models model id, or Copilot CLI model, then run
az login,gh auth login, orcopilot auth loginas needed.
- Check if
4. Summary
Print a short summary of what was created and any pending user actions. End with:
Try
/flow:research-requirements <issue-url-or-description>to kick off the workflow.
Out of scope
This skill does not:
- Commit anything (no
git add/git commit). - Run authentication flows (
gh auth login,az login,copilot auth login) - the user does those. - Install prerequisites (
jq,curl,gh,az,copilot) - the user installs them. - Overwrite existing
AGENTS.mdortracker.jsonfiles.
Errors
- If the project is not a git repo, the workflow skills assume git is in use. Suggest
git initif applicable, but do not run it automatically. - If a write fails (permissions, etc.), print the exact error and stop.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most project setup skills give in ~1.3k tokens
Counted across 1,553 of the 3,091 authors here whose files we hold, read 2026-09-06
- Write the configuration filein 36 of 1553
- Create the directory structurein 35 of 1553, across 33 files
- Verify the setupin 31 of 1553, across 28 files
- Run the setup scriptin 30 of 1553, across 29 files
- Pre-determine the required sample sizein 29 of 1553, across 12 files
- Check if the configuration already existsin 29 of 1553
- Document every testin 26 of 1553, across 10 files
- Start with a hypothesisin 26 of 1553, across 11 files
- Ask one question at a timein 22 of 1553
- Test a single variable per testin 21 of 1553, across 9 files
- Read product marketing context before asking questionsin 19 of 1553, across 8 files
- Do not peek and stop earlyin 18 of 1553, across 7 files
Said here and by no other author read
- Check if tracker.json exists
- Setup workflow-state labels for github
- Ask to set up cross-vendor review
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.