validate

Arc Validate — Configuration Validator

Validate the Maximus Loop project configuration by running deterministic CLI checks and providing project-aware advisories.

Announce: "I'll validate your Maximus Loop configuration and check for potential issues."

Phase 1: Run CLI Validation

1. Run maximus validate --json
2. Parse the JSON result with structure: { valid: true|false, checks: [{ name, status, message }], config_summary: { ... } | null }
3. If all checks pass (no failures): proceed to Phase 2 for advisories
4. If any checks fail: present failures clearly:

Validation Failures:
  ✗ [check.message for each failed check]

5. If user asks to fix issues, apply targeted fixes. Otherwise, report only.

Phase 2: Project-Aware Advisories

Only run if Phase 1 passes. Check these 5 things, only report mismatches:

1. Timeout vs project size

• Count files: find . -type f -not -path './node_modules/*' -not -path './.git/*' -not -path './dist/*' -not -path './build/*' -not -path './.next/*' -not -path './vendor/*' | wc -l
• Small (<100 files): 600s is fine, 900+ may be excessive
• Medium (100-500): 900s recommended
• Large (500+): 1200s recommended
• Report if timeout seems mismatched

2. Context files

• Check if context.files includes CLAUDE.md (if one exists in the project)
• Check if referenced context files actually exist

3. Escalation status

• If escalation is disabled but plan has tasks with complexity_level, note the mismatch

4. Commit prefix vs git log

• First check git rev-parse --is-inside-work-tree — skip if not a git repo or has no commits
• Run git log --oneline -5 and compare against git.commit_prefix
• If prefix doesn't match recent commit style, note it (informational only)

5. Plan health

• If plan has tasks, report completion status (X/Y completed, Z pending)
• If all tasks are complete, suggest: maximus archive → maximus clean → /arc:plan for next batch

6. Provider configuration

• If agent.provider is set in config, check it is a recognized value (claude or codex)
• If codex provider is configured, check that any per-task provider overrides also include a model field
• Note: the preflight check in the engine will catch CLI availability at runtime, but a config advisory here prevents surprises

Output format for advisories:

Advisories:
  • [advisory description]

If no advisories: skip this section entirely.

Phase 3: Present Findings

Valid with no advisories

✓ Configuration is valid

Configuration:
  Project:      [name]
  Model:        [model] (escalation: [status])
  Timeout:      [N]s
  Iterations:   [N]
  Auto-commit:  [yes/no] (prefix: "[prefix]")
  Auto-push:    [yes/no]

Ready to run: maximus run

Valid with advisories

✓ Configuration is valid (with advisories)

[Config summary as above]

Advisories:
  • [list]

Invalid

✗ Configuration has errors

Failures:
  ✗ [list]

[Advisories if any]

Fix the failures above, then re-run /arc:validate.

Constraints

• Read-only by default. Do NOT write files unless the user explicitly asks to fix something.
• Always distinguish CLI facts (deterministic) from advisories (skill observations). Never mix them.
• Short output when valid. Don't pad.
• Config schema reference: ${CLAUDE_PLUGIN_ROOT}/references/config-schema.md