korchasa/flowai-workflow-init

Initialize flowai-workflow Project

Overview

Scaffold a .flowai-workflow/ directory in the current project with workflow config, agent definitions, memory files, and HITL scripts. The skill analyzes the project to pre-fill settings, confirms them with the user, then runs flowai-workflow init non-interactively.

Instructions

Narration rule: before every step below, print one short sentence telling the user what you are about to do (what will be read, what will be written, what command will run). The goal is no surprises — the user should always know the next action before it happens. Do not bundle multiple steps into one announcement; narrate each one as you reach it.

1. Pre-check

Announce first, e.g.: "Checking whether .flowai-workflow/ already exists (read-only)."

• If .flowai-workflow/ already exists, inform the user and STOP. Do not overwrite without explicit confirmation. If the user confirms re-init, they must remove the directory manually first.

2. Analyze the project

Announce first, e.g.: "Reading deno.json / package.json / go.mod / Cargo.toml / pyproject.toml and git metadata to detect project settings. Nothing is modified."

Read manifest files to detect:

• PROJECT_NAME — from deno.json → name, package.json → name, go.mod → module path basename, Cargo.toml → [package] name, pyproject.toml → [project] name. Fallback: current directory name.
• DEFAULT_BRANCH — run git symbolic-ref refs/remotes/origin/HEAD, extract branch name after refs/remotes/origin/. Fallback: main.
• CHECK_CMD — single command that runs all project checks (format, lint, tests). Detect from: deno.json → tasks.check, package.json → scripts.check or scripts.test, or by stack: cargo test (Rust), go test ./... (Go), pytest (Python). If no check task but separate lint and test tasks exist, combine them: e.g. npm run lint && npm test.

3. Confirm with user

Announce first, e.g.: "Showing the detected values; I will wait for your confirmation before running init."

Present detected values and ask the user to confirm or edit:

I detected the following project settings:

  PROJECT_NAME:    <detected>
  DEFAULT_BRANCH:  <detected>
  CHECK_CMD:       <detected or empty>

Are these correct? If not, tell me what to change.

Wait for user response. Apply any corrections.

4. Run init

Announce first, e.g.: "Writing /tmp/flowai-init-answers.yaml (temp answers file) and running flowai-workflow init. This creates .flowai-workflow/ in the project root; no other paths are touched. The temp file is removed after init."

Write a temporary YAML answers file. The template uses two separate placeholders (TEST_CMD, LINT_CMD) — pass the single CHECK_CMD as both:

cat > /tmp/flowai-init-answers.yaml << 'EOF'
PROJECT_NAME: "<value>"
DEFAULT_BRANCH: "<value>"
TEST_CMD: "<CHECK_CMD value>"
LINT_CMD: "<CHECK_CMD value>"
EOF

flowai-workflow init --answers /tmp/flowai-init-answers.yaml
rm /tmp/flowai-init-answers.yaml

Pass through any user-provided flags (--template, --dry-run, --allow-dirty).

If init fails due to uncommitted changes, ask the user whether to pass --allow-dirty or commit first, then retry.

5. Post-init guidance

Announce first, e.g.: "Init finished. Printing follow-up steps; nothing else will be changed."

After successful scaffold, tell the user:

1. Review agents in .flowai-workflow/agents/agent-*.md and adapt prompts to your project conventions.
2. Review .flowai-workflow/workflow.yaml for workflow structure.
3. Run flowai-workflow run to execute the first workflow.

Available Templates

• sdlc-claude (default) — 6-agent SDLC workflow (PM → Architect → Tech Lead → Developer/QA loop → Tech Lead Review)