samvil-deploy

name	samvil-deploy
description	Deploy built app to Vercel/Railway/Coolify. Post-QA, pre-retro chain step.

samvil-deploy (ultra-thin)

Deploy a QA-verified app. The QA gate, per-solution_type deploy-target catalog (web-app / dashboard / game / mobile-app / automation), env-var validation, and build-artifact checks are aggregated by mcp__samvil_mcp__evaluate_deploy_target. Platform shell-out (vercel, railway, coolify, eas, gh-pages, crontab) stays here — host-bound, P8 graceful degradation. Full per-platform prose in SKILL.legacy.md.

Boot Sequence (INV-1)

1. Read project.state.json + project.seed.json + .env.example.
2. v3.2 Contract Layer — stage entry: mcp__samvil_mcp__save_event(session_id="<session_id>", event_type="deploy_started", stage="deploy", data="{}"). Best-effort; auto-claims evidence_posted subject="stage:deploy".

Step 1 — Aggregate readiness

mcp__samvil_mcp__evaluate_deploy_target(project_root=".", platform="")

Returns: qa_gate.{verdict,reason,next_action}, platforms[] {id,label,command,config_files,notes,recommended?}, selected_platform (caller arg → seed tech_stack.deploy → catalog recommended), env_vars.{required_keys,placeholder_keys,exists}, build_artifact.{checked_paths,present}, plus ready + blockers[]. On error: report ⚠ MCP unreachable, fall back to manual deploy guidance from SKILL.legacy.md (P8).

Step 2 — Branch on ready

ready: false — print blockers, stop:

• qa_gate.verdict != "pass" → refuse. Run /samvil:samvil-qa until PASS.
• env_vars.placeholder_keys non-empty → AskUserQuestion to fill values. Write .env / .env.production. Never hardcode secrets.
• build_artifact.present == false for non-manual → npm run build > .samvil/build.log 2>&1 (INV-2), then re-call evaluate_deploy_target.

ready: true — confirm selected_platform (silent if seed-pinned). For a different choice, re-call with platform="<id>" to recompute.

Step 3 — Deploy (host-bound)

cd ~/dev/<seed.name>
<selected_platform.command>   # substitute <seed.name> in the template

Failures: report once, no auto-retry. Per-platform recovery hints (Vercel cred, Railway nixpacks, Coolify Dockerfile, EAS keystore) live in SKILL.legacy.md.

automation deploy is recipe-driven, not a shell binary:

• cc-skill → call CronCreate with seed.core_flow.trigger.
• cron → write .samvil/crontab-template (template in SKILL.legacy.md), tell user crontab .samvil/crontab-template.
• serverless → generate serverless.yml (Python AWS Lambda) or vercel.json (Node cron), then run the command.
• manual → print run command + .env reminder.

Step 4 — Post-deploy

cat > .samvil/deploy-info.json <<EOF
{"platform":"<id>","url":"<URL or pending>","environment":"production",
 "deployed_at":"$(date -u +%Y-%m-%dT%H:%M:%SZ)","deploy_time_seconds":<int>}
EOF

Edit project.state.json: deploy_status: "deployed" (or "skipped") plus URL. Print one report (per-solution_type template in SKILL.legacy.md); minimum fields: platform, URL, environment, time.

Chain (mandatory)

Whether deployed or skipped:

1. Append Deploy section to .samvil/handoff.md.
2. Best-effort mcp__samvil_mcp__save_event(session_id="<session_id>", event_type="deploy_complete", stage="deploy", data='{"status":"<deployed|skipped>"}').
3. Immediately Invoke the Skill tool with skill: samvil-retro — no approval pause. Retro is read-only (no P10 irreversibility). Skipping breaks harness-feedback; v3.2 dogfood regressed here.

Anti-Patterns

QA-blocked deploy (gate refuses) · hardcoded secrets in .env (AskUserQuestion instead) · auto-retry on failure (report once, stop) · skipping samvil-retro chain. · AskUserQuestion 호출 포맷: questions 파라미터는 반드시 배열 — questions=["<질문>"]. 문자열 직접 전달 시 InputValidationError 발생.

Legacy reference

Full per-platform Korean prose, EAS pre-deploy checklist, post-deploy report templates per solution_type live in SKILL.legacy.md. Consult only when deploy itself is failing or being extended.