Release Safety

Release safety

Core principle: a backend release is valid only if the current production app keeps working after it deploys. Any change that breaks the current app version is an invalid release — regardless of whether a force update is planned.

1. Backend backward compatibility

Every backend release MUST work with BOTH the current production app and the new app — there's always a gap between backend deploy and app rollout.

Database changes:

• Only additive migrations per release (add columns, add tables).
• NEVER rename or drop columns in the same release as the feature using them.
• Sequence: add column → update reads/writes → remove old column in a later release.

API changes:

• Preserve existing API contracts (fields, response shapes, endpoints).
• Don't remove/break old fields or endpoints until the old app version is fully deprecated.

Code requirement: when introducing backward-compatible logic, add a comment explaining why it's needed, which app versions are affected, and when the old logic/fields can be removed.

2. Test current app against new backend

Before every release — including force-update releases — verify the current production app version works against the new backend. Run all critical user flows with the current app pointed at the new backend. If any critical flow fails: block the release, fix the backend, re-test. No exceptions.

3. Rollback plan

Every release requires a documented rollback plan before it ships, answering:

• Can the previous backend version be redeployed safely?
• If not, what's the forward-fix strategy?
• Who is the rollback owner (authority to trigger it)?
• What are the rollback conditions (error-rate threshold, critical failure types)?

Do not release if any of the above is undefined.

4. Config & data readiness

• NEVER assume staging DB, config, or seed data matches production — verify explicitly.
• Validate all critical config at startup; fail fast if missing/invalid — no silent fallbacks.
• Treat code readiness, DB migration readiness, and config readiness as three separate checks.
• Explicitly verify partner settings, category mappings, plan settings, and other business config.
• Keep a golden staging/UAT dataset covering critical flows and known edge cases.
• Don't rely on staging-only assumptions or seed data in production code paths.

5. Feature flags & observability

• Use feature flags for high-risk business-logic changes — do not gate on NODE_ENV.
• Add logs and alerts on every critical rule branch, fallback path, missing config, unexpected behavior.
• Post-release verification is mandatory for critical flows — smoke-test before AND after release.

6. Code comment policy (transition logic)

Add code comments only when needed for: backward compatibility (affected versions + when removable), migration safety (what state the DB must be in), release safety (deploy-order requirement), or temporary transition logic (TODO: remove after ).

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: kennguyen887
• Source: kennguyen887/agent-foundation
• License: MIT

Install and usage instructions live in the source repository linked above.