Deployless

Agent skills for moving a repository toward deployless delivery.

Deployless does not mean "no deployments." It means deployments become mechanical: every safe change merges to one mainline and moves through an automated path, while user-facing release is controlled separately at runtime with flags, kill switches, routing, config, or capability gates.

The result is a delivery model where:

• main is the single long-lived integration branch.
• Every merged change is production-deployable.
• CI gates protect the mainline before merge and deploy.
• Deployment from main is automated and repeatable.
• Release is controlled at runtime, not by long-lived branches.
• Risky data changes use expand-contract migrations.
• AI-generated PRs are scoped, owned, queued, and reviewed before they can pressure the mainline.
• Agent work moves through scope challenge, planning artifacts, test-first implementation, and independent review.
• Production behavior is observable, reversible, and covered by replay tests where practical.

What's included

Skill	Purpose
deployless-audit	Inspect a repo and create a deployless migration plan.
deployless-mainline	Move toward one long-lived mainline with merge gates.
deployless-release-controls	Add feature flags, kill switches, routing controls, or similar runtime gates.
deployless-production-path	Connect successful main commits to deployment, publication, or artifact promotion.
deployless-operational-safety	Add trace/replay testing, observability, rollback docs, and safe migration rules.
deployless-ai-pr-governance	Manage large AI PR inflows, queue pressure, and multi-agent work.

The skills are language-agnostic. They can be applied to services, frontends, CLIs, libraries, mobile apps, infrastructure repos, serverless functions, and monorepos.

Each skill follows the standard skill layout:

skills/
  skill-name/
    SKILL.md
    agents/openai.yaml

Install

Install the full package with the skills CLI:

npx skills add GuiBibeau/deployless

Install one skill from the package:

npx skills add GuiBibeau/deployless --skill deployless-audit

Install non-interactively for a specific agent:

npx skills add GuiBibeau/deployless --skill '*' --agent <agent-name> --yes

List the package contents without installing:

npx skills add GuiBibeau/deployless --list

Use

Run the skills in order. Start with the audit:

Use $deployless-audit on this repository.

Then continue through the sequence:

Use $deployless-mainline.
Use $deployless-release-controls.
Use $deployless-production-path.
Use $deployless-operational-safety.
Use $deployless-ai-pr-governance.

The audit should create or update docs/deployless-mainline-plan.md. Later skills append implementation details and safety guidance to that plan.

Registry

The package is installable from GitHub with npx skills add GuiBibeau/deployless. The public skills directory discovers packages from install and discovery activity; there is no separate publish command to run for this repository.

Recommended rollout

1. Audit the current branch, release, deployment, test, and observability model.
2. Make main the source of deployable truth.
3. Add runtime release controls before exposing unfinished behavior.
4. Automate deployment or publication from successful main commits.
5. Add rollback, tracing, replay tests, and migration safety.
6. Add AI PR intake and multi-agent work rules before PR volume overwhelms review.
7. Use a coordinator-led agent workflow: challenge unclear scope, publish ready slices, design alternatives for important interfaces, build with vertical test-first loops, and review standards plus spec before merge.

Do not enable automatic production deployment until the audit shows that tests, release controls, rollback, and migration safety are adequate.

Safety rules

• Do not remove review, tests, approvals, or compliance gates.
• Do not use long-lived environment branches as the target model.
• Do not conflate deployment with release.
• Do not commit production secrets or provider tokens.
• Do not perform destructive migrations without an expand-contract plan.
• Do not auto-merge AI PRs just because CI is green.
• Do not let multiple agents edit the same risky area without an owner.
• Do not let agents implement unclear work before scope, acceptance criteria, and test seams are resolved.
• Do not claim deployless delivery is complete while known safety blockers remain.

References

• Trunk-Based Development
• Atlassian: Trunk-based development

License

MIT