From charl
Plan, review, and implement safe production delivery for AI-authored Cloudflare changes and AI-powered products. Use when the user asks how to vibe code safely in production, ship AI-written Cloudflare features, classify risk tiers or blast radius, design previews, feature-flagged rollout, rollback, observability, or choose Cloudflare controls such as Workers, Pages, Flagship, AI Gateway, Access, D1, Durable Objects, Queues, Workflows, Browser Run, or Sandbox. Support both human-operated local workflows and autonomous agents running in sandboxes or remote VMs, and fail closed when required authority, verification, or rollback paths are missing.
How this skill is triggered — by the user, by Claude, or both
Slash command
/charl:charlThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Turn ambiguous "ship this with AI" work into a governed Cloudflare delivery
Turn ambiguous "ship this with AI" work into a governed Cloudflare delivery plan. Support both human-operated local workflows and autonomous agents running in sandboxes or remote VMs. Optimize for bounded blast radius, preview-first verification, controlled exposure, and fast rollback.
Choose one mode explicitly:
| Mode | Typical operator | Default stance |
|---|---|---|
human-local | Human and agent on the same workstation | Interactive review and deploy flow is allowed |
human-remote | Agent runs in a sandbox or remote VM, human reviews and promotes | Build, test, preview, and evidence generation are allowed; human owns final promotion |
autonomous-remote | Unattended agent in a sandbox or remote VM | Build, test, preview, and evidence generation are allowed; production promotion is constrained by the authority matrix |
autonomous-deployer | Unattended agent with explicit deploy rights and observability access | Allow promotion only when policy, verification, and rollback requirements are all satisfied |
If the mode is unknown, stop and report that as a blocker.
Require these capabilities before proceeding:
If any required capability is missing, stop with an explicit error and a handoff package. Do not improvise around missing permissions or missing verification.
| Tier | Typical scope | AI autonomy | Minimum release rule |
|---|---|---|---|
| 0 | Prototype, internal toy, throwaway experiment | High | Keep it away from production users and production data |
| 1 | Leaf-node UI, content transforms, internal tools, non-critical workflows | High | Require preview verification, exposure control, and rollback |
| 2 | Customer-facing logic with bounded data access | Medium | Review interfaces, auth path, dependencies, tests, observability, and ownership before rollout |
| 3 | Auth, billing, payments, shared schema, core architecture, compliance-sensitive systems | Low | Let AI assist, but do not let AI drive unsupervised merges or production release decisions |
Treat "leaf node" as a starting heuristic, not a guarantee. Re-check coupling to auth, billing, entitlements, analytics, email, rate limits, search, admin tools, and retention before declaring a change safe.
Apply these defaults unless the user or organization provides stricter policy:
| Mode | Tier 0 | Tier 1 | Tier 2 | Tier 3 |
|---|---|---|---|---|
human-local | Can build and release if requested | Can build and release with preview, evidence, and rollback | Can build, preview, and propose release; require explicit owner review before production | Assist only; do not drive release |
human-remote | Can build and release if human owns promotion | Can build, preview, and release with explicit human approval | Can build and preview; require explicit human promotion | Assist only; do not drive release |
autonomous-remote | Can build and release only if preview, evidence, and rollback exist | Can build, test, preview, and dark-launch if policy allows | Can build, test, preview, and stage; do not promote to broad production without explicit deploy authority | Assist only; do not drive release |
autonomous-deployer | Can build and release if verification and rollback pass | Can build, preview, dark-launch, and progressively expose if policy allows | Can promote only with explicit deploy authority, production observability, rollback access, and a named owner | Assist only; do not drive release |
When the current request exceeds the allowed action for its mode and tier, stop and hand off instead of partially deploying.
Require these artifacts for any non-trivial change:
If one of these is missing, create it before implementation.
human-local, prefer local development first. In remote modes, use preview
builds and remote bindings intentionally. Protect sensitive previews with
Access. Pin preview_urls in Wrangler so dashboard toggles do not drift on
the next deploy. Review Workers Builds token scope instead of blindly keeping
the auto-created token. If the Worker uses Durable Objects, do not assume
Preview URLs are available.default gateway pattern, gateway
retries, and Unified Billing or ZDR where the data policy requires it. If ZDR
is a hard requirement, verify provider support and gateway settings and fail
closed otherwise.If blocked, return the blocker, the missing capability, and the exact handoff artifacts the next operator needs.
For low-risk changes, review behavior harder than code. For high-risk changes, review behavior and interfaces harder than code. For Tier 3 changes, review the code too.
Use this structure unless the user asks for a different format:
## Execution mode
[human-local | human-remote | autonomous-remote | autonomous-deployer]
## Capability check
- [Available capabilities]
- [Missing capabilities]
- [Whether work can continue]
## Risk tier
[Tier and justification]
## Blast radius
[Systems, data, and side-effects touched]
## Required Cloudflare controls
- [Build and preview]
- [Exposure control]
- [Runtime and state]
- [Secrets and access]
- [Observability]
## Verification gates
- [Automated checks]
- [Manual checks]
- [Adversarial or abuse-path check]
## Rollout and rollback
- [Release shape]
- [Rollback limits]
## Open risks
- [Remaining concerns or blockers]
## Handoff
- [What the next human or agent needs if this mode cannot complete the task]
Provides CDSS development patterns for drug interaction checking, dose validation, clinical scoring (NEWS2, qSOFA), and alert classification integrated into EMR workflows.
npx claudepluginhub charl-kruger/skills --plugin charl