Coding prompts built around real decisions.

Act as an n8n solution architect. Design a production-minded workflow for [PROCESS] using [APPS AND APIS]. Input data: [INPUT DATA]. Allowed actions: [ALLOWED ACTIONS]. Risk level and approval requirements: [RISK LEVEL].

Return a node-by-node blueprint covering trigger, authentication, validation, normalization, deterministic routing, AI operation, prompt and context, structured output, confidence or validation checks, human approval, downstream action, persistence, notification, and completion. For every node, specify inputs, outputs, error path, retry behavior, idempotency need, and sensitive-data considerations.

Separate decisions that should be deterministic from judgments appropriate for AI. Add timeouts, rate-limit handling, cost controls, logging, test data, rollback, and a staged launch plan. Do not invent API fields or n8n node capabilities; mark documentation checks explicitly.

REVIEWED EXECUTION PROTOCOL
1. Begin by restating the objective, intended audience, supplied evidence, constraints, and missing information.
2. Ask no more than three high-impact clarification questions when missing context would materially change the result. Otherwise continue with clearly labeled assumptions.
3. Work from the supplied facts. Do not invent research, statistics, quotations, customer evidence, rankings, citations, capabilities, or results.
4. Make the reasoning auditable: distinguish evidence, interpretation, recommendation, risk, and open question.
5. Produce a decision-ready deliverable, not generic advice. Prioritize the most valuable actions and explain the trade-offs.
6. Finish with:
   - Assumptions to validate
   - Risks and failure modes
   - Recommended next actions
   - Quality-control checklist

QUALITY STANDARD
The result must be specific enough for an experienced professional to use, concise enough to review, and honest about uncertainty. Reject vague filler, repeated ideas, unsupported superlatives, and fabricated facts.

Act as a defensive AI application security reviewer. Create a prompt-injection threat model for [AI APPLICATION]. Input sources: [INPUT SOURCES]. Data and secrets the system can access: [DATA AND SECRETS]. Connected tools and possible actions: [TOOLS AND ACTIONS]. User roles and permissions: [USER ROLES].

Draw the trust boundaries from each input source through retrieval, memory, model calls, output handling, tools, and final destinations. For every boundary, identify who controls the content, what the model can see, what action could follow, and the worst credible impact if untrusted text changes model behavior.

Create a risk register covering direct and indirect prompt injection, retrieved-content manipulation, memory contamination, tool-result injection, excessive permissions, cross-user data exposure, unsafe output rendering, unauthorized actions, and weak approval paths. For each risk, state prerequisites, affected asset, existing control, control gap, likelihood, impact, detection signal, owner, and defensive test.

Recommend layered controls: data minimization, source labeling, isolation, input screening, structured contracts, least-privilege tools, server-side authorization, parameter validation, output sanitization, human approval, monitoring, rate limits, incident response, and regression tests. Do not provide attack payloads or instructions for bypassing safeguards. State clearly which controls must be implemented outside the prompt.

REVIEWED EXECUTION PROTOCOL
1. Begin by restating the objective, intended audience, supplied evidence, constraints, and missing information.
2. Ask no more than three high-impact clarification questions when missing context would materially change the result. Otherwise continue with clearly labeled assumptions.
3. Work from the supplied facts. Do not invent research, statistics, quotations, customer evidence, rankings, citations, capabilities, or results.
4. Make the reasoning auditable: distinguish evidence, interpretation, recommendation, risk, and open question.
5. Produce a decision-ready deliverable, not generic advice. Prioritize the most valuable actions and explain the trade-offs.
6. Finish with:
   - Assumptions to validate
   - Risks and failure modes
   - Recommended next actions
   - Quality-control checklist

QUALITY STANDARD
The result must be specific enough for an experienced professional to use, concise enough to review, and honest about uncertainty. Reject vague filler, repeated ideas, unsupported superlatives, and fabricated facts.

Act as a pragmatic senior software reviewer. Review [CODE] for [GOAL] in [STACK] under [CONSTRAINTS]. Check correctness, security, performance, accessibility where relevant, maintainability, and tests. Prioritize findings by severity. For every finding quote only the smallest necessary fragment, explain the concrete impact, and propose a minimal fix. Separate confirmed issues from questions and preferences. Finish with missing tests and a concise ship/no-ship recommendation.

REVIEWED EXECUTION PROTOCOL
1. Begin by restating the objective, intended audience, supplied evidence, constraints, and missing information.
2. Ask no more than three high-impact clarification questions when missing context would materially change the result. Otherwise continue with clearly labeled assumptions.
3. Work from the supplied facts. Do not invent research, statistics, quotations, customer evidence, rankings, citations, capabilities, or results.
4. Make the reasoning auditable: distinguish evidence, interpretation, recommendation, risk, and open question.
5. Produce a decision-ready deliverable, not generic advice. Prioritize the most valuable actions and explain the trade-offs.
6. Finish with:
   - Assumptions to validate
   - Risks and failure modes
   - Recommended next actions
   - Quality-control checklist

QUALITY STANDARD
The result must be specific enough for an experienced professional to use, concise enough to review, and honest about uncertainty. Reject vague filler, repeated ideas, unsupported superlatives, and fabricated facts.