Limitations

Limitations

8-habit-ai-dev is workflow discipline packaged as markdown skills. It helps agents and humans ask better questions, preserve evidence, review AI-generated work, and hand off decisions. It does not execute work by itself.

Important

Treat this plugin as guidance, not authority. A skill can remind you to verify, review, stage, or monitor; it cannot make a production action safe on its own.

Where It Helps

Situation	Useful Starting Point	Why
Unclear problem or unfamiliar domain	/research	Forces source-backed investigation before requirements
Feature planning	/requirements	Defines scope, success criteria, and EARS-style behavior before implementation
Architecture choices	/design	Surfaces trade-offs and human decision points
AI-generated implementation	/review-ai	Produces a verdict with evidence, severity, and verification expectations
Pre-merge or pre-release uncertainty	/cross-verify	Checks the work across all 8 habits and Body/Mind/Heart/Spirit dimensions
Operational finding or incident closure	/operational-state, /deploy-guide, /post-mortem	Separates observation, action, rollout, and learning

What It Does Not Do

Not Provided	Use Instead
Runtime policy enforcement	Companion tooling such as claude-governance or host-specific gates
Irreversible-action authorization	Human approval and the deployment/change-management system you already trust
Compliance certification or legal advice	Qualified compliance/legal review and auditable control evidence
Cloud, Kubernetes, Docker, or database mutation	Your normal infrastructure tooling, runbooks, and staged rollout process
Secret scanning or credential handling	Dedicated security scanners and secret-management systems
Dynamic sub-agent orchestration engine	A separate adapter or orchestration product
Claude hook feature parity inside Codex	Codex-native adapters or explicit manual checks

Platform Boundaries

Claude Code and Codex both receive the same markdown skills. Claude Code may also run Claude-specific hooks from hooks/. If Codex invokes this package's SessionStart hook, the hook returns JSON additional context rather than raw markdown; broader hook behavior remains host-specific. Other agents can use the repo as markdown instructions when they can load AGENTS.md, skills/RESOLVER.md, and the relevant SKILL.md.

The shared source of truth remains skills/*/SKILL.md. Packaging, hooks, banners, and marketplace behavior are host-specific presentation layers.

Evidence Expectations

When a change affects install behavior, generated discovery metadata, skill routing, docs, or release trust, prove it from the surface the user will touch:

• run the repository validators,
• check generated catalog freshness when metadata changes,
• verify installed plugin state for Claude Code or Codex when packaging changes,
• link the exact docs or wiki page affected,
• state what was not tested.

Do not use secrets, customer-sensitive raw data, or private incident transcripts as proof.

See Also

• Architecture
• Installation
• Troubleshooting
• Contributing to Wiki