init

Arc Init

Announce: "I'll validate your current setup, analyze the project, and configure Maximus Loop."

## Task API is MANDATORY — 4 separate tasks

You MUST call TaskCreate to create a NEW, SEPARATE task for EACH phase. This skill produces exactly 4 tasks:

1. TaskCreate → "Detect & validate existing setup"
2. TaskCreate → "Analyze project structure"
3. TaskCreate → "Configure Maximus settings"
4. TaskCreate → "Validate & handoff"

Do NOT reuse a task from a previous phase. Each BEGIN block creates a fresh task. Each END block completes that phase's task. The user sees 4 checkboxes — one per phase.

If you skip Task API calls or reuse tasks across phases, this skill has FAILED.

---

Phase 1: Detect & Validate

BEGIN: Call TaskCreate (NEW task, do not reuse previous) with subject "Detect & validate existing setup", description "Run maximus validate --json and determine current state", activeForm "Detecting existing setup". Then call TaskUpdate with status in_progress for that task.

Step 1: Run Validation

Run maximus validate --json

This is your FIRST action after creating the task. Do NOT read files, explore, or run anything else before this.

Step 2: Parse Validation Output

Parse the JSON output and check the directory check status. This determines which state path to follow.

State A — No Maximus Setup (directory check has status: "fail")

No .maximus/ directory exists.

Announce:

No Maximus setup found. I'll analyze the project and create a tailored configuration.

Proceed to Phase 2.

State B — Existing Setup with Errors (directory exists, other checks fail)

.maximus/ exists but validation shows failures.

Announce:

Existing Maximus setup found with errors:
  [list each check where status is "fail" with its message]

Proceed to Phase 2. CRITICAL: Do NOT run maximus init in Phase 3 — it refuses to re-initialize existing directories. In Phase 3, apply targeted fixes instead.

State C — Valid Setup (valid: true)

All checks pass. Show the config_summary from the JSON:

Valid Maximus setup found:
  Project:      [config_summary.project_name]
  Model:        [config_summary.default_model] (escalation: [enabled/disabled])
  Timeout:      [config_summary.timeout]s
  Iterations:   [config_summary.max_iterations]
  Auto-commit:  [yes/no] (prefix: "[config_summary.commit_prefix]")
  Auto-push:    [yes/no]

Use AskUserQuestion: "Would you like to change anything?" with options:

• "No, this is good"
• "Yes, I want to change settings"

If user selects "No, this is good":

• Skip to Phase 4 (skip Phases 2 and 3 entirely)

If user selects "Yes, I want to change settings":

• Note what settings to change
• Proceed to Phase 2

END: Call TaskUpdate with status completed for the Phase 1 task.

---

Phase 2: Analyze

BEGIN: Call TaskCreate (NEW task, do not reuse previous) with subject "Analyze project structure", description "Read package.json, git log, and count files to determine 3 config values", activeForm "Analyzing project structure". Then call TaskUpdate with status in_progress for that task.

Step 1: Extract Project Name

Read package.json (or Cargo.toml, go.mod, pyproject.toml for other languages).

Extract the name field value.

Step 2: Detect Commit Prefix

Run git log --oneline -10

Look for consistent prefixes in commit messages (e.g., "feat:", "fix:", "maximus:").

If a prefix pattern is found, use it. If no consistent pattern exists, default to "maximus:".

Step 3: Calculate Timeout

Run this exact command in Bash:

find . -type f -not -path './node_modules/*' -not -path './.git/*' -not -path './dist/*' -not -path './build/*' -not -path './.next/*' -not -path './vendor/*' | wc -l

Calculate timeout based on file count:

• Less than 100 files → 600
• 100-500 files → 900
• More than 500 files → 1200

Step 4: Print Analysis Summary

Print:

Project Analysis:
  Name:           [project-name from Step 1]
  Timeout:        [N]s (based on ~[X] files)
  Commit prefix:  "[detected-prefix from Step 2]"

Steps 1-3 are the ONLY reads in this phase. Do NOT read, explore, or analyze: - README, tsconfig, or any config files beyond package.json - Framework or dependency details - Project architecture or integrations - Test configuration or CI setup - Any file not explicitly listed in Steps 1-3

END: Call TaskUpdate with status completed for the Phase 2 task.

---

Phase 3: Configure

BEGIN: Call TaskCreate (NEW task, do not reuse previous) with subject "Configure Maximus settings", description "Run maximus init, then edit 3 config values from Phase 2", activeForm "Configuring Maximus settings". Then call TaskUpdate with status in_progress for that task.

Step 1: Initialize Directory

For State A only (no existing setup):

Run maximus init with no flags and no arguments.

Never use mkdir. Never add flags — maximus init --help will execute init instead of showing help.

For State B (existing setup with errors):

Skip this step entirely. The .maximus/ directory already exists, and maximus init refuses to re-initialize existing directories.

Step 2: Read Default Configuration

Read .maximus/config.yml to see the current/default values.

Step 3: Update Project Name

Use Edit tool to change project_name to the value from Phase 2, Step 1.

Step 4: Update Timeout

Use Edit tool to change the timeout value under the agent: section to the value from Phase 2, Step 3 (600, 900, or 1200).

Step 5: Update Commit Prefix

Use Edit tool to change commit_prefix under the git: section to the value from Phase 2, Step 2.

CONSTRAINT: These are the ONLY 3 values you modify. Do NOT change any other fields. Do NOT add fields. Do NOT use Write tool on config.yml — only use Edit.

Step 6: Initialize Plan File

Write .maximus/plan.json with:

{
  "version": "1.0.0",
  "tasks": []
}

If the file already exists, read it first then overwrite it.

Step 7: User Confirmation (BLOCKING)

Present the 3 modified values:

Configuration:
  Project name:   [value from Step 3]
  Timeout:        [value from Step 4]s
  Commit prefix:  "[value from Step 5]"

Use AskUserQuestion: "Does this configuration look correct?" with options:

• "Yes, looks good"
• "No, I want to change something"

If user selects "No, I want to change something":

• Ask what to change
• Apply changes using Edit tool
• Re-confirm with another AskUserQuestion
• Repeat until user approves

Do NOT proceed past this step until user selects "Yes, looks good".

END: Call TaskUpdate with status completed for the Phase 3 task after user confirms.

---

Phase 4: Validate & Handoff

BEGIN: Call TaskCreate (NEW task, do not reuse previous) with subject "Validate & handoff", description "Run final validation and present next steps", activeForm "Validating final configuration". Then call TaskUpdate with status in_progress for that task.

Step 1: Final Validation

Run maximus validate --json

Step 2: Handle Validation Result

If valid: true:

Print:

Maximus Loop is configured for [project-name]

Configuration saved to:
  .maximus/config.yml — Engine settings
  .maximus/plan.json — Empty task list (ready for planning)
  .maximus/progress.md — Iteration tracker

Next Steps:
  1. Commit setup:
     git add .maximus/ .gitignore
     git commit -m "[prefix] Initialize Maximus Loop"
  2. Create your first task plan:
     Run /arc:plan to design tasks for your feature
  3. After each run completes:
     maximus archive  — save results to .maximus/archive/
     maximus clean    — reset runtime state for the next batch

If invalid:

Show failures from the validation output.

Attempt to fix the issues (max 2 retries):

1. Identify the failing check
2. Apply targeted fix using Edit tool
3. Run maximus validate --json again
4. If still invalid, repeat once more (max 2 total attempts)

If still invalid after 2 attempts:

• Show the persistent failures
• Ask the user for help: "I've attempted to fix these issues but validation still fails. Could you help me understand what needs to change?"

END: Call TaskUpdate with status completed for the Phase 4 task.

---

Alternate Paths

This skill has three possible execution paths determined in Phase 1, Step 2:

State A: No Existing Setup

Trigger: directory check has status: "fail" in validation JSON

Path: Phase 1 → Phase 2 → Phase 3 (with maximus init in Step 1) → Phase 4

Description: Fresh initialization. Creates .maximus/ directory, analyzes project, configures settings, validates.

State B: Existing Setup with Errors

Trigger: .maximus/ exists but other validation checks fail

Path: Phase 1 → Phase 2 → Phase 3 (skip maximus init in Step 1) → Phase 4

Description: Repair mode. Skips maximus init (refuses to re-initialize), analyzes project, applies targeted fixes to existing config, validates.

Critical difference: Step 1 is skipped because maximus init will fail on existing directories.

State C: Valid Existing Setup

Trigger: valid: true in validation JSON and user selects "No, this is good"

Path: Phase 1 → Phase 4 (Phases 2 and 3 skipped entirely)

Description: Configuration already valid. Skips analysis and configuration, jumps directly to validation and handoff.

Critical difference: If user wants changes, treat as State B and proceed through all phases.