A Claude Code skill for zero-surprise production deployments.
Born from real incidents where "it's deployed" turned out to mean "only 3 of 18 changed files actually made it to the server."
Enforces a 7-step closed-loop every time you deploy code to a production server:
PRE-CHECK --> BACKUP --> DEPLOY --> VERIFY FILES --> VERIFY FUNCTION --> MONITOR LOGS --> NOTIFY
- SHA-256 file integrity verification -- checksums every file, local vs remote, diff must be zero
- Sensitive file protection --
.env, private keys, SQLite databases automatically excluded from deployment - Cross-OS permission fix -- solves the macOS-to-Linux ownership problem (
501:staff->www:www) - Database migration coordination -- migration order table based on change type
- Rollback with security review -- warns if rollback would re-introduce fixed vulnerabilities
- Log monitoring with secret redaction -- strips passwords and tokens from terminal output
- Multi-server canary deployment -- deploy to least critical server first
- CI/CD integration mapping -- 7 steps map directly to pipeline stages
- 10 anti-patterns with correct alternatives
- 5 real-world incident case studies
- Any web stack: PHP, Node.js, Python, Go, Ruby, etc.
- Any deployment method: tar, rsync, scp, CI/CD
- Any server: bare metal, VPS, cloud instances
- Any OS combination: macOS/Windows/Linux to Linux
Copy the skill to your Claude Code skills directory:
# Create the skill directory
mkdir -p ~/.claude/skills/deploy-verify
# Download SKILL.md
curl -o ~/.claude/skills/deploy-verify/SKILL.md \
https://raw.githubusercontent.com/Web3ok/deploy-verify-skill/main/SKILL.mdOr clone the repo:
git clone https://github.com/Web3ok/deploy-verify-skill.git ~/.claude/skills/deploy-verifyThe skill triggers automatically when you say things like:
- "deploy to server"
- "sync to production"
- "push to live"
- "upload to server"
Or invoke it explicitly:
/deploy-verify
Most deployment skills focus on CI/CD pipelines (merge PR, wait for CI, deploy). But for teams deploying via tar/rsync/scp to bare metal or VPS servers, there's a critical gap: nobody verifies the files actually arrived.
This skill fills that gap with battle-tested practices from real production incidents:
- 18-file silent desync -- Incremental deploys left 18 files outdated. Admin panel missing entire sections.
- macOS ownership = Linux 500 -- Files with owner
501:staffand600permissions. Entire site down 30 min. - OPcache served old code -- PHP-FPM
reloaddidn't clear OPcache. New code on disk, old code executed. - CDN 520 from restart -- PHP-FPM restart behind Cloudflare caused 520 for all users.
- .env overwritten -- tar deployed local
.envover production credentials. Wrong database connected.
Issues and PRs welcome. If you've had a deployment horror story that this skill should cover, open an issue.
MIT