cto-loop

cto-loop

Semi-autonomous variant of full-auto — same pipeline, plus two operator gates and an interactive tech-stack intake. Use when a tech lead is available to review the plan upfront and the running app at the end.

This skill is a thin overlay on full-auto. All shared logic (transcript analysis, project-type detection, audit, masterplan assembly, builder dispatch, version search, docker-compose deploy, QA loop, fixer subagent) lives in ../full-auto/ and is referenced from there. Only the gate flows + the interactive overlays live here.

When to use

• A transcript + a tech lead who wants to review at two points
• Greenfield builds where the operator wants to choose the stack interactively (instead of accepting full-auto's autonomous defaults)
• Brownfield builds where the operator wants to confirm the audit before the masterplan locks in

When NOT to use

• No human in the loop → use full-auto
• One-line ask, no transcript → use sonzai-sdk wizard
• Spec + plan only, no build → use sonzai-sdk wizard

Pipeline overlay

The pipeline mirrors full-auto's exactly, with two gates inserted and two phases swapped for interactive variants:

Phase	full-auto behavior	cto-loop overlay
0-pre Notify-setup	../full-auto/notify-setup.md (autonomous)	notify-setup.md (interactive — asks if missing)
Pre-flight	drift check, docker check, key check	identical
0a Transcript	../full-auto/transcript-analysis.md	identical
0b Project type	../full-auto/project-type-detection.md	identical
0c-G Greenfield stack	../full-auto/tech-stack-derivation.md (autonomous)	tech-stack-intake.md (interactive 7Q)
0c-B Brownfield audit	../full-auto/brownfield-audit.md (autonomous)	brownfield-audit.md (detect-and-confirm)
0d Wizard answers	../full-auto/answer-derivation.md	identical
1 Masterplan	../full-auto/masterplan-assembly.md	identical
🚪 Gate A	(none)	masterplan-gate.md
2 Build	../full-auto/builder-dispatch.md + subagent prompts in ../full-auto/subagent-prompts/	identical
3 Deploy + QA	../full-auto/local-deploy.md + ../full-auto/qa-loop.md	identical
🚪 Gate B	(none — auto-fixer instead)	cto-review-gate.md
4 Iterate	../full-auto/feedback-iteration.md (auto-fixer on QA failure)	feedback-iteration.md (operator feedback drives fixer)
5 Final report	../full-auto/final-report.md.template	identical (with operator approval reflected)

Hard rules (overlay-specific)

1. Two gates are non-optional. No skip-flag. If you want autonomous, use full-auto.
2. Interactive overrides override autonomous. When cto-loop is active, Phase 0c-G uses tech-stack-intake.md (interactive 7Q) instead of full-auto's tech-stack-derivation.md; brownfield-audit prints findings for row-by-row confirmation; feedback-iteration loops on operator feedback rather than QA-failure auto-fixing.

(All other hard rules — always-search, no remote git/deploy, no tenant names, etc. — are inherited from ../full-auto/SKILL.md. They apply identically in this mode.)

Files (cto-loop-specific only)

File	Purpose
notify-setup.md	Phase 0-pre: interactive ask for Gmail/Slack if not cached
pipeline.md	Overlay pipeline with gates inserted
tech-stack-intake.md	Interactive 7-question variant of tech-stack-derivation
brownfield-audit.md	Detect-and-confirm wrapper around full-auto's audit
masterplan-gate.md	Gate A flow (approve / edit / reject the masterplan)
cto-review-gate.md	Gate B flow (live app review)
feedback-iteration.md	Operator-feedback-driven fixer loop (bounded 5 cycles)

Everything else is referenced via ../full-auto/<file>.md.

Versioning

Ships with the sonzai-sdk plugin in sonzai-claude-skill. See repo CHANGELOG.