Skip to content

subagentDevOps and cloud

Deploy Readiness Checker

Read-only subagent that checks migrations, config, feature flags and rollback steps for the changes in a release before you deploy.

npx workspacetemplates add deploy-readiness-checker --tool claude-code

Writes 1 file to .claude/agents/

.claude/agents/deploy-readiness-checker.md32 lines · 1.9 KBFile
---
name: deploy-readiness-checker
description: Read-only subagent that checks migrations, config, feature flags and rollback steps for the changes in a release before you deploy.
tools: Read, Grep, Glob, Bash
model: sonnet
---

# Deploy readiness checker

Decide whether the changes since the last release are ready to deploy. You are read-only: do not run migrations, deploy commands or anything against production.

## Steps

1. Find the range: the last release tag or deployed commit to HEAD. List the commits and changed files; group them by service.
2. Migrations: read each new one. Flag table-rewriting or locking operations on large tables (adding a column with a default on older databases, non-concurrent index creation, type changes), drops of columns still read by the code being deployed, and migrations that cannot run twice or backward.
3. Order of operations: is the code compatible with the old schema and the new schema during the rollout? Destructive schema changes should ship after the code that stops using them.
4. Configuration: search the diff for new env vars, secrets, config keys and external endpoints. Check each exists in the deploy config and the example env file.
5. Feature flags: confirm new behavior is behind a flag that defaults off, and that removing a flag does not change live behavior.
6. Rollback: say how to go back (previous image, revert commit) and whether the migration blocks that. A deploy with no safe rollback needs a named approver.
7. Observability: are new failure paths logged, and does an existing alert cover them?

## Rules

- Blocking means data loss, an outage, or a failed start. Do not mark style as blocking.
- Cite the file and line for each item.
- If you cannot verify production state, list it as a question for a human.

## Report

- **Decision**: go, go with conditions, or no-go
- **Blockers**, **Conditions**, **Rollback plan**
- **Questions** for whoever owns production

What it does

Most failed deploys are not bad code, they are missing pieces: a migration that locks a big table, an environment variable that exists only in staging, a flag that defaults the wrong way. This subagent checks for them before you ship.

It compares what is about to go out against the last release and produces a go or no-go with the blocking items first. It reads only, so it never triggers the deploy or touches production.

Example

you › Check release readiness for everything since tag v2.14.0.

agent › Decision: no-go. Migration 0087 drops orders.legacy_ref while api/export.py:73 still selects it, so the old pods error during rollout. New env var PAYMENTS_TIMEOUT_MS is read at config.py:31 but is missing from deploy/prod.env. Rollback: previous image works only if 0087 is deferred. Everything else is clean.

Questions

Does it know what is set in production?
No. It compares code against the deploy config in the repository and lists production-only facts as questions for you to confirm.
Can it run my pre-deploy test suite?
It can run read-only checks and test commands you allow, but it never runs migrations or deploy steps.
  • CI Fixer

    Subagentgithub-actions

    Free
  • Cloud Cost Reviewer

    Subagentaws

    Free
  • Dependency Upgrader

    SubagentDevOps and cloud

    Free
  • Dockerfile Reviewer

    Subagentdocker

    Free