swain-init

Project Onboarding

One-time setup for adopting swain in a project. This skill is not idempotent — it migrates files and installs tools. For per-session health checks, use swain-doctor.

Preflight

Before any phase, run the preflight script to gather environment state. This single call replaces all inline check blocks — phases below read from the JSON output instead of running shell commands.

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
PREFLIGHT_SCRIPT="$(

Store PREFLIGHT_JSON for use in all phases below. Every decision references a field from this JSON — do not run additional check commands unless performing a mutation.

Phase 0: Already-initialized detection

Read marker.action from the preflight JSON.

• "delegate" — same major version. Tell the user:

  Project already initialized (swain marker.release_version, init vmarker.last_version). Delegating to swain-session.

  Skip to Phase 7 (Session Start) below. Do not run Phases 1–6.

• "upgrade" — newer major version available. Tell the user:

  Project was initialized with swain marker.last_release_version (init vmarker.last_version). Current: marker.release_version (init vmarker.current_version). Consider running /swain update to pick up new features. Starting session.

  Skip to Phase 7 (Session Start) below. Do not re-run onboarding — upgrades are handled by swain-update, not swain-init.

• "onboard" — no marker found. Proceed with full onboarding (Phases 1–6).

Phase 1: CLAUDE.md → AGENTS.md migration

Goal: establish the @AGENTS.md include pattern so project instructions live in AGENTS.md (which works across Claude Code, GitHub, and other tools that read AGENTS.md natively).

Read migration.state from the preflight JSON.

If "fresh"

Create both files:

• CLAUDE.md: @AGENTS.md
• AGENTS.md: # AGENTS.md (empty — governance added in Phase 5)

If "migrated"

Skip to Phase 2.

If "standard"

1. Copy CLAUDE.md content to AGENTS.md (preserve everything).
2. If CLAUDE.md contains a <!-- swain governance --> block, strip it from the AGENTS.md copy — it will be re-added cleanly in Phase 5.
3. Replace CLAUDE.md with @AGENTS.md.

Tell the user:

Migrated your CLAUDE.md content to AGENTS.md and replaced CLAUDE.md with @AGENTS.md. Your existing instructions are preserved — Claude Code reads AGENTS.md via the include directive.

If "split"

Both files have content. Ask the user:

Both CLAUDE.md and AGENTS.md have content. How should I proceed?

1. Merge — append CLAUDE.md content to the end of AGENTS.md, then replace CLAUDE.md with @AGENTS.md
2. Keep AGENTS.md — discard CLAUDE.md content, replace CLAUDE.md with @AGENTS.md
3. Abort — leave both files as-is, skip migration

If merge: append CLAUDE.md content (minus any <!-- swain governance --> block) to AGENTS.md, replace CLAUDE.md with @AGENTS.md.

Phase 2: Verify dependencies

Step 2.1 — uv

Read uv.available from the preflight JSON.

If false, install:

If installation fails, tell the user:

uv installation failed. You can install it manually (https://docs.astral.sh/uv/getting-started/installation/) — swain scripts require uv for Python execution.

Skip the rest of Phase 2 on failure (don't block init on uv, but warn that scripts will not function without it).

Step 2.2 — Vendored tk

Read tk.path and tk.healthy from the preflight JSON.

If tk.path is null or tk.healthy is false, tell the user:

The vendored tk script was not found or is broken. This usually means the swain-do skill was not fully installed. Try running /swain update to reinstall skills.

Step 2.3 — Migrate from beads (if applicable)

Read beads.exists and beads.has_backup from the preflight JSON.

If beads.exists is true and beads.has_backup is true, offer migration:

Found existing .beads/ data. Migrate tasks to tk? This will convert .beads/backup/issues.jsonl to .tickets/ markdown files.

If user agrees, run migration:

Tell the user the results and that .beads/ can be removed after verification.

If beads.exists is false, skip. tk creates .tickets/ on first tk create.

Step 2.4 — Operator bin/ symlinks (SPEC-214, ADR-019)

Read bin_manifests from the preflight JSON. For each entry, create bin/ symlinks:

Tell the user which operator commands are now available in bin/.

If bin_manifests is empty, skip silently.

Phase 3: Pre-commit security hooks

Goal: configure pre-commit hooks for secret scanning so credentials are caught before they enter git history. Default scanner is gitleaks; additional scanners (TruffleHog, Trivy, OSV-Scanner) are opt-in.

Step 3.1 — Check for existing config

Read precommit.config_exists from the preflight JSON.

If true: Present the current .pre-commit-config.yaml content and ask:

Found existing .pre-commit-config.yaml. How should I proceed?

1. Merge — add swain's gitleaks hook alongside your existing hooks
2. Skip — leave pre-commit config unchanged
3. Replace — overwrite with swain's default config (your existing hooks will be lost)

If user chooses Skip, skip to Phase 4.

If false: Proceed to Step 3.2.

Step 3.2 — Install pre-commit framework

Read precommit.framework from the preflight JSON.

If false, install:

If uv is unavailable or installation fails, warn:

pre-commit framework not available. You can install it manually (uv tool install pre-commit or pip install pre-commit). Skipping hook setup.

Skip to Phase 4 if pre-commit cannot be installed.

Step 3.3 — Create or update .pre-commit-config.yaml

The default config enables gitleaks:

If the user requested additional scanners (via --scanner flags or when asked), add their hooks:

TruffleHog (opt-in):

Trivy (opt-in):

OSV-Scanner (opt-in):

Write the config file. If merging with an existing config, append the new repo entries to the existing repos: list.

Step 3.4 — Install hooks

Run pre-commit install to activate the hooks.

Step 3.5 — Update swain.settings.json

Read the existing swain.settings.json (if any) and add the sync.scanners key:

Set enabled: true for any scanners the user opted into. Merge with existing settings — do not overwrite other keys.

Tell the user:

Pre-commit hooks configured with gitleaks (default). Scanner settings saved to swain.settings.json. To enable additional scanners later, edit swain.settings.json and re-run /swain-init.

Phase 4: Superpowers companion

Goal: offer to install obra/superpowers if it is not already present.

Step 4.1 — Detect superpowers

Read superpowers.installed from the preflight JSON.

If true, report "Superpowers: already installed" and skip to Phase 4.4.

Step 4.2 — Offer installation

Ask the user:

Superpowers (obra/superpowers) is not installed. It provides TDD, brainstorming, plan writing, and verification skills that swain chains into during implementation and design work.

Install superpowers now? (yes/no)

If the user says no, note "Superpowers: skipped" and continue to Phase 4.4. They can always install later: npx skills add obra/superpowers.

Step 4.3 — Install

If the install succeeds, tell the user:

Superpowers installed. Brainstorming, TDD, plan writing, and verification skills are now available.

If it fails, warn:

Superpowers installation failed. You can retry manually: npx skills add obra/superpowers

Continue to Phase 4.4 regardless.

Step 4.4 — Tmux

Read tmux.installed from the preflight JSON.

If true, report "tmux: already installed" and continue to Phase 4.5.

If false, ask the user:

tmux is not installed. swain uses tmux for tab naming when available. It is optional — swain works without it, but session tab-naming will be unavailable.

Install tmux now? (yes/no)

If yes:

If the install succeeds, tell the user:

tmux installed. Workspace layout and tab naming features are now available.

If the install fails, warn:

tmux installation failed. You can install it manually: brew install tmux

If no, note "tmux: skipped" and continue to Phase 4.5.

Phase 4.5: Shell launcher

Goal: offer to install a swain shell function so the user can launch swain with a single command. Templates are stored per-runtime, per-shell in templates/launchers/{runtime}/swain.{shell} (relative to this skill's directory) — inspect them to see exactly what gets added. Supported runtimes are defined in ADR-017.

Step 4.5.1 — Detect shell

Read launcher.shell from the preflight JSON.

Supported shells: zsh, bash, fish. If the shell is not in this list, tell the user:

Shell launcher templates are available for zsh, bash, and fish. Your shell (launcher.shell) is not yet supported — skipping launcher setup.

Skip to Phase 5.

Step 4.5.2 — Check for existing launcher

Read launcher.already_installed from the preflight JSON.

If true, report "Shell launcher: already installed" and skip to Phase 5. Do not modify existing functions.

Step 4.5.3 — Detect runtimes

Read launcher.runtimes from the preflight JSON.

If the array is empty, tell the user:

No supported agentic CLI runtimes found (checked: claude, gemini, codex, copilot, crush). Install one first, then re-run /swain-init.

Skip to Phase 5.

Step 4.5.4 — Select runtime

• One runtime found: Offer it directly.
• Multiple runtimes found: Present a numbered list and ask which one to use. Default to claude if available.

Read launcher.template_dir from the preflight JSON. Construct the template path:

Step 4.5.5 — Show template and offer installation

Read the template file content and present it to the user:

Shell launcher — Add a swain command to your shell?

Detected runtime: [runtime name]. Here's what will be added to <rc-file>:

<template content>

Install? (yes/no)

For Crush templates, add a note: "Crush has partial support — it cannot accept an initial prompt, so session initialization relies on AGENTS.md auto-invoke directives."

Step 4.5.6 — Install

If the user accepts, append the template content to the rc file (read launcher.rc_file from preflight JSON, e.g. cat "$TEMPLATE_FILE" >> "$RC_FILE").

Tell the user:

Shell launcher installed. Run source <rc-file> (or restart your shell) to activate the swain command.

If the user declines, note "Shell launcher: skipped" and continue to Phase 5.

Phase 5: Swain governance

Goal: add swain's routing and governance rules to AGENTS.md.

Step 5.1 — Check for existing governance

Read governance.installed from the preflight JSON.

If true, governance is already installed. Tell the user and skip to Phase 5.5.

Step 5.2 — Ask permission

Ask the user:

Ready to add swain governance rules to AGENTS.md. These rules:

• Route artifact requests (specs, stories, ADRs, etc.) to swain-design
• Route task tracking to swain-do (using tk)
• Enforce the pre-implementation protocol (plan before code)
• Prefer swain skills over built-in alternatives

Add governance rules to AGENTS.md? (yes/no)

If no, skip to Phase 5.5.

Step 5.3 — Inject governance

Read the canonical governance content from swain-doctor/references/AGENTS.content.md (search .claude/skills, .agents/skills, and skills directories for the file). Append the full contents of that file to AGENTS.md.

Tell the user:

Governance rules added to AGENTS.md. These ensure swain skills are routable and conventions are enforced. You can customize anything outside the <!-- swain governance --> markers.

Phase 5.5: README seeding and artifact proposals (SPEC-207)

Goal: ensure every swain project has a README, and offer to bootstrap artifacts from it when the artifact tree is empty.

Step 5.5.1 — Check for README

Read readme.exists from the preflight JSON.

Step 5.5.2 — Seed README if missing

If readme.exists is false, determine the project's context using readme.has_code and readme.has_artifacts from the preflight JSON:

• No code, no artifacts — Interview the operator: "What does this project do?" Write the README from their answer.
• Code exists, no artifacts — Infer project purpose from code (read entry points, package.json/pyproject.toml/go.mod, etc.). Present a draft README to the operator for editing.
• Artifacts exist, no README — Compile from Active Visions, Designs, Journeys, and Personas. Present a draft to the operator for editing.

Present the draft to the operator. They can approve, edit, or skip. If they skip, note "README: skipped" in the summary and swain-doctor will flag it on future sessions.

Step 5.5.3 — Propose seed artifacts from README

Read readme.active_count from the preflight JSON.

If readme.active_count < 3 and README.md exists, read the README and extract intent claims using semantic analysis. Propose seed artifacts:

• Vision — from the README's description of what the project does and why.
• Personas — from who the README addresses and what problems it describes.
• Journeys — from usage flows, examples, or "getting started" paths.
• Designs — from architectural or structural claims.

Present each proposal individually. The operator approves, edits, or rejects each one. Approved artifacts are created via swain-design. Rejected proposals are silently dropped.

Semantic extraction: Read the entire README as prose. No convention-based sections, no operator-placed markers. Any claim in the README is a potential intent source — install instructions, feature descriptions, behavioral claims, architectural statements.

Phase 6: Finalize

Step 6.1 — Create .agents directory

Create .agents/ if it does not exist (mkdir -p .agents). This directory is used by swain-do for configuration and by swain-design scripts for logs.

Step 6.1.1 — Bootstrap .agents/bin/ (ADR-019)

Create .agents/bin/ and populate it with symlinks for all agent-facing scripts in the skill tree:

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
AGENTS_BIN="$REPO_ROOT/.agents/bin"
SKILLS_ROOT="$REPO_ROOT/.agents/skills"
mkdir -p "$AGENTS_BIN"
OPERATOR_SCRIPTS="swain swain-box"
for skill_scripts_dir in "$SKILLS_ROOT"/*/scripts; do
  [ -d "$skill_scripts_dir" ] || continue
  for script in "$skill_scripts_dir"/*; do
    [ -f "$script" ] && [ -x "$script" ] || continue
    script_name="$(basename "$script")"
    case "$script_name" in test-*) continue ;; esac
    echo " $OPERATOR_SCRIPTS " | grep -q " $script_name " && continue
    rel_path="$(python3 -c "import os,sys; print(os.path.relpath(sys.argv[1], sys.argv[2]))" "$script" "$AGENTS_BIN" 2>/dev/null)" || continue
    ln -sf "$rel_path" "$AGENTS_BIN/$script_name"
  done
done

Add .agents/bin/ and .agents/session.json to .gitignore if not already present (consumer projects should not track these).

Step 6.2 — Run swain-doctor

Invoke the swain-doctor skill. This validates .tickets/ health, checks stale locks, removes legacy skill directories, and ensures governance is correctly installed.

Step 6.3 — Onboarding

Invoke the swain-help skill in onboarding mode to give the user a guided orientation of what they just installed.

Step 6.4 — Write .swain/init.json marker

After all onboarding phases complete, write the .swain/init.json marker file. Read marker.current_version from the preflight JSON for the skill version and marker.release_version for the release version.

If .swain/init.json already exists (partial re-init), read it and append to the history array. Otherwise create a new file:

{
  "history": [
    {
      "version": "4.0.0",
      "release": "v0.29.0-alpha",
      "timestamp": "2026-03-26T18:30:00Z",
      "action": "init"
    }
  ]
}

For upgrades (future use by swain-update), append an entry with "action": "upgrade" instead.

Write the file and ensure .swain/ is in .gitignore (it's project-local state, not shared).

Step 6.5 — Summary

Report what was done:

swain init complete.

• CLAUDE.md → @AGENTS.md include pattern: [done/skipped/already set up]
• tk (ticket) verified: [done/not found]
• Beads migration: [done/skipped/no beads found]
• Pre-commit security hooks: [done/skipped/already configured]
• Superpowers: [installed/skipped/already present]
• tmux: [installed/skipped/already present]
• Shell launcher: [installed (runtime)/skipped/already present/no runtime found/unsupported shell]
• Swain governance in AGENTS.md: [done/skipped/already present]
• README: [seeded/already present/skipped]
• Artifact proposals from README: [N proposed, M accepted/skipped/not applicable]
• Init marker: written (.swain/init.json)

Step 6.6 — Start session

After successful onboarding, proceed to Phase 7 (Session Start) below.

Phase 7: Session Start (ADR-023)

This phase runs every time — both after fresh onboarding (Phase 6) and on the already-initialized fast path (Phase 0). It replaces the former swain-session startup sequence.

Step 7.1 — Fast greeting

Run the fast greeting script. It calls the session preflight internally for all read-only state, then applies lightweight mutations (tab naming, lock cleanup). No subprocess chain — one preflight pass.

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
bash "$REPO_ROOT/.agents/bin/swain-session-greeting.sh" --json

The greeting emits structured JSON:

{
  "greeting": true,
  "branch": "trunk",
  "dirty": false,
  "isolated": false,
  "bookmark": "Left off implementing the bootstrap script",
  "focus": "VISION-001",
  "tab": "project @ branch",
  "warnings": []
}

The session preflight (called internally by the greeting) also gathers previous session state. To access it directly:

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
PREFLIGHT_JSON=$( bash "$REPO_ROOT/.agents/bin/swain-session-preflight.sh" --repo-root "$REPO_ROOT" 2>/dev/null )

After receiving the greeting JSON:

1. Present the greeting to the operator — branch, dirty state, bookmark (if any), focus lane (if any), and warnings.

2. If bookmark is not null, display it:

   Resuming session — Last time: {bookmark}

3. If isolated is false, do not create a worktree now — worktree creation is deferred to swain-do task dispatch (SPEC-195).

If $TMUX is NOT set (detected by absence of tab in the JSON), check whether tmux is installed:

• tmux not installed: Offer to install it (brew install tmux).
• tmux installed but not in a session: Show: [note] Not in a tmux session — session tab and pane features unavailable

Step 7.2 — Session state init

Read prev_session from the session preflight JSON (or call the preflight directly if the greeting didn't expose it). If prev_session.exists is true, display its focus lane, walkaway note, and decision count. Ask the operator: "Continue previous session or start fresh?"

If starting fresh (or no previous session):

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
bash "$REPO_ROOT/.agents/bin/swain-session-state.sh" init \
  --focus "<FOCUS-ID>" \
  --session-roadmap "$(pwd)/SESSION-ROADMAP.md" \
  --repo-root "$REPO_ROOT"

Step 7.3 — Focus lane

The focus lane scopes recommendations to a single vision or initiative. It is set when the operator decides what to work on.

If the greeting JSON included a focus value: Confirm with the operator:

Focus lane: {focus}. Continue with this focus, or change?

If no focus is set: Ask the operator what they want to work on. Resolve names to artifact IDs:

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
bash "$REPO_ROOT/.agents/bin/chart.sh" --ids --flat 2>/dev/null | grep -i "<name>"

If exactly one match, use it. If multiple, ask the operator to clarify. If no match, tell the operator and offer to create one.

Set the focus:

bash "$REPO_ROOT/.agents/bin/swain-focus.sh" set <RESOLVED-ID>

Display the focus artifact context:

bash "$REPO_ROOT/.agents/bin/artifact-context.sh" <RESOLVED-ID> 2>/dev/null

The focus lane is stored in .agents/session.json under the focus_lane key and persists across the session.

Step 7.4 — Session purpose text

When the operator launches with free text (e.g., swain new bug about timestamps), the launcher exports SWAIN_PURPOSE and — for runtimes that accept an initial prompt — also passes it inline as /swain-init Session purpose: new bug about timestamps.

The greeting script (swain-session-greeting.sh) reads $SWAIN_PURPOSE and writes the bookmark deterministically (SPEC-297). The purpose field in the greeting JSON surfaces the captured text.

When the greeting JSON's purpose field is non-null:

• Display it to the operator: **Session purpose:** <text>.

Do not re-parse the initial prompt or call swain-bookmark.sh yourself — the greeting already did both. The inline prompt text is for display context only; the env var is the source of truth.

Worktree / branch changes

When an agent enters a worktree or switches branches, re-run the bootstrap with --path to update the tab name:

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
bash "$REPO_ROOT/.agents/bin/swain-session-bootstrap.sh" --path "$NEW_WORKDIR" --skip-worktree --auto

Re-running init

If the user runs /swain-init on a project that's already set up, Phase 0 reads the preflight JSON's marker.action field and skips to Phase 7 (Session Start) — no onboarding phases run, no interactive prompts appear. This lets users build muscle memory around /swain-init as a single entry point.

To force re-onboarding, delete .swain/init.json and re-run.