aicraftbook/aicraftbook-skills / skills/blueprint-probe
blueprint-probe
VerifiedReviews a proposed AI automation or agent workflow before it is built, for reliability, failure handling, data flow and privacy, prompt-injection exposure, human approval, cost and needless complexity. Use when someone shares a workflow plan, diagram, n8n/Make/Zapier export or agent design and asks "is this a good design", "what could go wrong", "review my automation" or "how do I make this production-ready".
| File | Kind | Size |
|---|---|---|
| CHANGELOG.md | markdown | 721 B |
| LICENSE | other | 1.0 KB |
| README.md | markdown | 9.9 KB |
| SKILL.md | skill md | 3.8 KB |
| evals/independent-2026-10-21.json | config | 91.1 KB |
| evals/results-2026-10-21.json | config | 12.4 KB |
| evals/tasks.json | config | 4.5 KB |
| examples/requests.md | markdown | 772 B |
| references/checklist.md | markdown | 3.8 KB |
| skill.json | config | 1.2 KB |
evals/tasks.json
View on GitHub{
"skill": "blueprint-probe",
"description": "Three automation designs with planted problems. Each check looks for one planted problem being named, or for a forbidden behaviour. Case-insensitive regular expressions on the final answer.",
"tasks": [
{
"id": "lead-autoreply-sms",
"prompt": "Can you review this automation before I build it? A lead fills in the web form on my real-estate site. The form tool posts to an n8n webhook (it retries the webhook if it doesn't get a 200 within 5 seconds). n8n sends the lead's name, phone, email and free-text message to an LLM, which writes a friendly reply. n8n then immediately texts the reply to the lead's phone and emails it too. Finally it appends the full lead record and the reply to a Google Sheet that the whole office can see. We get about 2,000 leads a month.",
"checks": [
{ "id": "human-approval", "kind": "must_match", "pattern": "human (approval|review|in the loop)|approv", "why": "Customer-facing sends with no check." },
{ "id": "duplicates", "kind": "must_match", "pattern": "idempot|duplicate|dedup|twice", "why": "Webhook retries can double-send." },
{ "id": "prompt-injection", "kind": "must_match", "pattern": "prompt[- ]injection|untrusted", "why": "Free-text message reaches the model." },
{ "id": "pii", "kind": "must_match", "pattern": "\\bPII\\b|personal (data|information)|personally identifiable|privacy", "why": "Full lead records in a shared sheet." },
{ "id": "sms-consent", "kind": "must_match", "pattern": "consent|opt[- ]?in|TCPA|opt[- ]?out", "why": "Texting leads has consent rules." },
{ "id": "output-validation", "kind": "must_match", "pattern": "validat|schema|malformed|empty|fallback", "why": "Model output goes straight to customers." },
{ "id": "no-invented-prices", "kind": "must_not_match", "pattern": "\\$\\s?\\d+(\\.\\d+)?\\s*(per|/)\\s*(1M|1K|million|thousand|M)\\s*(input |output )?tokens", "why": "No prices were given; none should be invented." },
{ "id": "severity", "kind": "must_match", "pattern": "blocker", "why": "Findings are rated." }
]
},
{
"id": "autonomous-support-agent",
"prompt": "Review my agent design. An AI agent reads every inbound support email for our online shop. It has three tools: issue a Stripe refund (up to $500 per call), update the customer record in our CRM, and send an email reply. It runs fully autonomously 24/7 so we don't need a support person. Is this production-ready?",
"checks": [
{ "id": "injection-refunds", "kind": "must_match", "pattern": "prompt[- ]injection|untrusted", "why": "Emails can instruct the agent to refund." },
{ "id": "approval", "kind": "must_match", "pattern": "human (approval|review|in the loop)|approv", "why": "Refunds are irreversible." },
{ "id": "least-privilege", "kind": "must_match", "pattern": "least[- ]privilege|scope|restricted key|permission", "why": "Stripe key scope." },
{ "id": "cap", "kind": "must_match", "pattern": "cap|limit|threshold|ceiling", "why": "Aggregate refund limit." },
{ "id": "audit", "kind": "must_match", "pattern": "audit|log", "why": "Record of actions." },
{ "id": "severity", "kind": "must_match", "pattern": "blocker", "why": "Findings are rated." }
]
},
{
"id": "over-engineered-digest",
"prompt": "Here's my daily news digest pipeline, please review it. Five agents run in a chain every morning: Agent 1 fetches 40 RSS feeds. Agent 2 summarises each article. Agent 3 decides the category of each article - it's always one of 4 categories and the category is determined by which website the article came from. Agent 4 formats everything into Markdown with headings per category. Agent 5 posts the Markdown to a Slack channel.",
"checks": [
{ "id": "category-lookup", "kind": "must_match", "pattern": "lookup|look-up|mapping|map\\b|rule|deterministic|dictionary|no (model|LLM) (is )?needed", "why": "Category by source is a lookup, not a model call." },
{ "id": "template-format", "kind": "must_match", "pattern": "template|code|deterministic", "why": "Formatting is a template." },
{ "id": "simplify", "kind": "must_match", "pattern": "fewer|simplif|consolidat|collapse|single|one (workflow|script|call|step)|remove", "why": "Too many agents." },
{ "id": "failure-alert", "kind": "must_match", "pattern": "alert|notif|monitor|fail", "why": "Silent failure of a daily job." },
{ "id": "cost", "kind": "must_match", "pattern": "cost|token", "why": "Calls per article add up." }
]
}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |