solo-deploy（单独部署）

/deploy

Deploy the project to its hosting platform. Reads the stack template YAML (templates/stacks/{stack}.yaml) for exact deploy config (platform, CLI tools, infra tier, CI/CD, monitoring), detects installed CLI tools, sets up database and environment, pushes code, and verifies deployment is live.

Live Context

• Branch:!git branch --show-current 2>/dev/null
• Uncommitted:!git status --short 2>/dev/null | wc -l | tr -d ' ' files
• Last commit:!git log --oneline -1 2>/dev/null

References

• templates/principles/dev-principles.md — CI/CD, secrets, DNS, shared infra rules
• dev-principles-my.md (if exists) — personal infra extensions (auth, tunnels, monitoring)
• templates/stacks/*.yaml — Stack templates with deploy, infra, ci_cd, monitoring fields

Paths are relative to the skill's plugin root. Search for these files via Glob if not found at expected location.

When to use

After /build has completed all tasks (build stage is complete). This is the deployment engine.

Pipeline: /build → /deploy → /review

MCP Tools (use if available)

• session_search(query) — find how similar projects were deployed before
• project_code_search(query, project) — find deployment patterns across projects
• codegraph_query(query) — check project dependencies and stack

If MCP tools are not available, fall back to Glob + Grep + Read.

Pre-flight Checks

0. Check if deploy is needed

Read docs/workflow.md (if exists). If it contains "No deploy" or "no deploy stage" or "ship = commit":

• Output: "Deploy skipped — workflow.md says no deploy stage (CLI/local project)."
• Output <solo:done/> immediately.
• Do NOT proceed with any deploy steps.

Also: if project has NO deploy infrastructure (no Dockerfile, no deploy.sh, no wrangler.toml, no vercel.json, no sst.config.ts) AND workflow.md exists:

• Same: skip deploy, output done.

1. Verify build is complete (optional)

• If pipeline state tracking exists (.solo/states/ directory), check .solo/states/build.
• If .solo/states/ exists but build marker is missing: warn "Build may not be complete. Consider running /build first."
• If .solo/states/ does not exist: skip this check and proceed with deployment.

2. Detect available CLI tools

Run in parallel — detect what's installed locally:

vercel --version 2>/dev/null && echo "VERCEL_CLI=yes" || echo "VERCEL_CLI=no"
wrangler --version 2>/dev/null && echo "WRANGLER_CLI=yes" || echo "WRANGLER_CLI=no"
npx supabase --version 2>/dev/null && echo "SUPABASE_CLI=yes" || echo "SUPABASE_CLI=no"
fly version 2>/dev/null && echo "FLY_CLI=yes" || echo "FLY_CLI=no"
sst version 2>/dev/null && echo "SST_CLI=yes" || echo "SST_CLI=no"
gh --version 2>/dev/null && echo "GH_CLI=yes" || echo "GH_CLI=no"
cargo --version 2>/dev/null && echo "CARGO_CLI=yes" || echo "CARGO_CLI=no"

Record which tools are available. Use them directly when found — do NOT npx if CLI is already installed globally.

3. Load project context (parallel reads)

• CLAUDE.md — stack name, architecture, deploy platform
• docs/prd.md — product requirements, deployment notes
• docs/workflow.md — CI/CD policy (if exists)
• package.json or pyproject.toml — dependencies, scripts
• wrangler.toml, sst.config.ts, vercel.json — platform configs (if exist)
• docs/plan/*/plan.md — active plan (look for deploy-related phases/tasks)

Plan-driven deploy: If the active plan contains deploy phases or tasks (e.g. "deploy Python backend to VPS", "run deploy.sh", "set up Docker on server"), treat those as primary deploy instructions. The plan knows the project-specific deploy targets that the generic stack YAML may not cover. Execute plan deploy tasks in addition to (or instead of) the standard platform deploy below.

3b. Detect project type

Classify the project to select the right deploy strategy:

For CLI/TUI/library projects: skip web deploy steps (Vercel, CF, etc.) — go directly to package registry deploy (crates.io, PyPI, npm).

3c. Cross-compile compatibility check (Rust/native projects)

When: Cargo.toml exists AND CI/CD will build on a different architecture (e.g., GitHub Actions x86_64 vs local Apple Silicon).

# Check for native/system dependencies that may fail cross-compile
grep -r 'links\s*=' Cargo.toml 2>/dev/null         # C library bindings
grep -r 'build\s*=' Cargo.toml 2>/dev/null          # build.rs (may compile C code)
grep -rn 'sys' Cargo.toml 2>/dev/null | grep -v '#' # *-sys crates (native bindings)

Known cross-compile risks:

• ort-sys / onnxruntime-sys — no prebuilt binaries for all targets, requires local build
• ffmpeg-sys / ffmpeg-next — needs FFmpeg installed on CI runner
• openssl-sys — use rustls instead for pure-Rust TLS
• Any *-sys crate — check if prebuilt binaries exist for CI target architecture

If risky deps found:

1. Check if CI matrix includes the required target (.github/workflows/*.yml)
2. Recommend adding cross-compile toolchain to CI, OR
3. Recommend replacing native dep with pure-Rust alternative, OR
4. Add the dep to CI install step: apt-get install / brew install
5. Report in deploy summary which deps may block CI

4. Read stack template YAML

Extract the stack name from CLAUDE.md (look for stack: field or tech stack section).

Read the stack template to get exact deploy configuration:

Search order (first found wins):

1. templates/stacks/{stack}.yaml — relative to this skill's plugin root
2. .solo/stacks/{stack}.yaml — user's local overrides (from /init)
3. Search via Glob for **/stacks/{stack}.yaml in project or parent directories

Extract these fields from the YAML:

• deploy — target platform(s): vercel, cloudflare_workers, cloudflare_pages, docker, hetzner, app_store, play_store, local
• deploy_cli — CLI tools and their use cases (e.g. vercel (local preview, env vars, promote))
• infra — infrastructure tool and tier (e.g. sst (sst.config.ts) — Tier 1)
• ci_cd — CI/CD system (e.g. github_actions)
• monitoring — monitoring/analytics (e.g. posthog)
• database / orm — database and ORM if any (affects migration step)
• storage — storage services if any (R2, D1, KV, etc.)
• notes — stack-specific deployment notes

Use the YAML values as the source of truth for all deploy decisions below. The YAML overrides the fallback tier matrix.

5. Detect platform (fallback if no YAML)

If stack YAML was not found, use this fallback matrix:

If $ARGUMENTS specifies a platform, use that instead of auto-detection or YAML.

Auto-deploy platforms (from YAML deploy field or fallback):

• vercel / cloudflare_pages — auto-deploy on push. Push to GitHub is sufficient if project is already linked. Only run manual deploy for initial setup.
• cloudflare_workers — wrangler deploy needed (no git-based auto-deploy for Workers).
• docker / hetzner — manual deploy via SSH or Pulumi.

Deployment Steps

Step 1. Git — Clean State + Push

git status
git log --oneline -5

If dirty, commit remaining changes:

git add -A
git commit -m "chore: pre-deploy cleanup"

Ensure remote exists and push:

git remote -v
git push origin main

If no remote, create GitHub repo:

gh repo create {project-name} --private --source=. --push

For platforms with auto-deploy (Vercel, CF Pages): pushing to main triggers deployment automatically. Skip manual deploy commands if project is already linked.

Step 2. Database Setup

Supabase (if supabase/ dir or Supabase deps detected):

# If supabase CLI available:
supabase db push          # apply migrations
supabase gen types --lang=typescript --local > db/types.ts  # optional: regenerate types

If no CLI: guide user to Supabase dashboard for migration.

Drizzle ORM (if drizzle.config.ts exists):

npx drizzle-kit push      # push schema to database
npx drizzle-kit generate  # generate migration files (if needed)

D1 (Cloudflare) (if wrangler.toml has D1 bindings):

wrangler d1 migrations apply {db-name}

If database is not configured yet, list what's needed and continue — don't block on it.

Step 3. Environment Variables

Read .env.example or .env.local.example to identify required variables.

Generate platform-specific instructions:

Vercel:

# If vercel CLI is available and project is linked:
vercel env ls  # show current env vars

# Guide user:
echo "Set env vars: vercel env add VARIABLE_NAME"
echo "Or via dashboard: https://vercel.com/[team]/[project]/settings/environment-variables"

Cloudflare:

wrangler secret put VARIABLE_NAME  # interactive prompt for value
# Or in wrangler.toml [vars] section for non-secret values

Do NOT create or modify .env files with real secrets. List what's needed, let user set values.

Step 4. Platform Deploy

See references/platform-commands.md for full deploy, env var, and log commands per platform.

Use the detected platform from pre-flight checks:

• Vercel — vercel link → vercel (preview) → vercel --prod
• Cloudflare Workers — wrangler deploy
• Cloudflare Pages — wrangler pages deploy./out
• SST — sst deploy --stage prod

For auto-deploy platforms (Vercel, CF Pages): git push origin main is sufficient if already linked.

Step 4b. Rust/Crates.io Deploy

Detect: Cargo.toml exists at project root AND cargo CLI is available.

Determine deploy target from Cargo.toml:

• If [lib] section exists or crate is a library → crates.io publish
• If only [[bin]] → binary release (GitHub Releases)
• If both → do both (publish crate + attach binary to release)

Pre-publish checklist:

# 1. Verify tests + clippy pass
cargo test
cargo clippy --all-targets -- -D warnings

# 2. Check Cargo.toml has required fields for crates.io
grep -E '^(name|version|description|license|repository)' Cargo.toml

# 3. Dry run publish
cargo publish --dry-run

Required Cargo.toml fields for crates.io (fail without them):

• name, version, description, license (or license-file), repository

If missing fields: add them, commit, then proceed.

Publish to crates.io:

# Check if already published at this version
CRATE_NAME=$(grep '^name' Cargo.toml | head -1 | sed 's/.*= *"//' | sed 's/".*//')
CRATE_VER=$(grep '^version' Cargo.toml | head -1 | sed 's/.*= *"//' | sed 's/".*//')

# Publish (requires `cargo login` — if not authenticated, note in report and skip)
cargo publish

If publish fails:

• already uploaded → version already exists, bump version in Cargo.toml
• not logged in → note in report: "Run cargo login with crates.io API token"
• dependency not on crates.io → check path/git dependencies, replace with crates.io versions

Binary release (optional, for CLI tools):

# Build release binary
cargo build --release

# Create GitHub release with binary attached
BINARY_NAME=$(grep '^name' Cargo.toml | head -1 | sed 's/.*= *"//' | sed 's/".*//')
gh release create "v${CRATE_VER}" --title "v${CRATE_VER}" --generate-notes \
  "target/release/${BINARY_NAME}#${BINARY_NAME}-$(uname -s)-$(uname -m)"

Verify crate on crates.io:

# Wait a few seconds for index to update, then check
curl -s "https://crates.io/api/v1/crates/${CRATE_NAME}/${CRATE_VER}" | head -100

Step 5. Verify Deployment

After deployment, verify it actually works:

STATUS=$(curl -s -o /dev/null -w "%{http_code}" https://{deployment-url})
BODY=$(curl -s https://{deployment-url} | head -200)

If HTTP status is not 200: check env vars, fix, redeploy. See references/platform-commands.md for platform-specific env var and error resolution commands.

Do NOT output <solo:done/> until the live URL returns HTTP 200 and page loads without errors. If you cannot fix the issue, output <solo:redo/> to go back to build. Output pipeline signals ONLY if .solo/states/ directory exists.

Step 6. Post-Deploy Log Monitoring

After verifying HTTP 200, tail production logs to catch runtime errors. See references/platform-commands.md for platform-specific log commands and what to look for.

What to do with log errors:

• Env var missing → fix with platform CLI (see Step 3), redeploy
• DB connection error → check connection string, IP allowlist
• Runtime crash / unhandled error → if .solo/states/ exists, output <solo:redo/> to go back to build with fix; otherwise fix and redeploy
• No errors in 30 lines of logs → proceed to report

If logs show zero traffic (fresh deploy), make a few test requests:

curl -s https://{deployment-url}/           # homepage
curl -s https://{deployment-url}/api/health  # API health (if exists)

Then re-check logs for any errors triggered by these requests.

Step 7. Post-Deploy Report

Deployment: {project-name}

  Platform:  {platform}
  URL:       {deployment-url}
  Branch:    main
  Commit:    {sha}

  Done:
    - [x] Code pushed to GitHub
    - [x] Deployed to {platform}
    - [x] Database migrations applied (or N/A)

  Manual steps remaining:
    - [ ] Set environment variables (listed above)
    - [ ] Custom domain (optional)
    - [ ] PostHog / analytics setup (optional)

  Next: /review — final quality gate

Completion

Signal completion

If .solo/states/ directory exists, output this exact tag ONCE and ONLY ONCE — the pipeline detects the first occurrence:

<solo:done/>

Do NOT repeat the signal tag anywhere else in the response. One occurrence only. If .solo/states/ directory does not exist, skip the signal tag.

Error Handling

CLI not found

Cause: Platform CLI not installed. Fix: Install the specific CLI: npm i -g vercel, npm i -g wrangler, brew install supabase/tap/supabase.

Deploy fails — build error

Cause: Build works locally but fails on platform (different Node version, missing env vars). Fix: Check platform build logs. Ensure engines in package.json matches platform. Set missing env vars.

Database connection fails

Cause: DATABASE_URL not set or network rules block connection. Fix: Check connection string, platform's DB dashboard, IP allowlist.

Git push rejected

Cause: Remote has diverged. Fix: git pull --rebase origin main, resolve conflicts, push again.

Verification Gate

Before reporting "deployment successful":

1. Run curl -s -o /dev/null -w "%{http_code}" against the deployment URL.
2. Verify HTTP 200 (not 404, 500, or redirect loop).
3. Check the actual page content matches expectations (not a blank page or error).
4. Only then report the deployment as successful.

Never say "deployment should be live" — verify it IS live.

Pipeline / Autonomous Mode

When running in a pipeline (--print, no human watching):

1. NEVER call AskUserQuestion — responses are lost between iterations, causing infinite retry loops (observed: 5 wasted iterations on OpenWok deploy)
2. Make autonomous deploy decisions:

- No deploy CLI installed → install via brew/npm if possible, otherwise create Dockerfile +.dockerignore, push code, mark "Ready for manual deploy" - No auth/credentials → create all deploy configs, push to GitHub, write state file with auth instructions, signal done - Multiple platform options → pick based on: stack YAML > Next.js/React → Vercel, Python → Cloudflare Workers (MVP) or Hetzner (prod), static → Cloudflare Pages

1. Skip over blockers — browser login, manual approval, missing API keys → note in report, continue
2. Always signal completion — partial deploy (configs created + code pushed) is better than infinite retry. Output <solo:done/> even if actual deployment requires manual auth step.

Critical Rules

1. Use installed CLIs — detect vercel, wrangler, supabase, fly, sst before falling back to npx.
2. Auto-deploy aware — if platform auto-deploys on push, just push. Don't run manual deploy commands unnecessarily.
3. NEVER commit secrets — no.env files with real values, no API keys in code.
4. Preview before production — deploy preview first, verify, then promote to prod.
5. Check build locally first — pnpm build / uv build (or equivalent) before deploying.
6. Check production logs — always tail logs after deploy, catch runtime errors before declaring success.
7. Report all URLs — deployment URL + platform dashboard links.
8. Infrastructure in repo — prefer sst.config.ts or wrangler.toml over manual dashboard config.
9. Verify before claiming done — HTTP 200 from the live URL + clean logs, not just "deploy command succeeded".