Prompt Cards

Create a source scoring rubric: domain trust, author expertise, citations, recency, and conflicts of interest. Output a weighted scoring model and how to tune it per domain.

Design a query planning module that expands a user question into search operators, sub-queries, and entity lists. Include multilingual support and query diversification strategies.

Draft a compliance-first plan: robots.txt handling, site terms, user-agent identification, rate limiting, and respectful crawling. Include an ethics checklist and escalation rules.

Create a selection matrix for available APIs: search/SERP providers, content extraction, crawling, and anti-bot friendly approaches. Compare cost, rate limits, coverage, and TOS constraints.

Design an end-to-end architecture for a research bot that uses web search APIs, page fetching, and summarization. Include modules, data flow, storage, caching, and citation strategy.

Create a 90-day plan to build a high-capability agent system using MCP and A2A: milestones, architecture decisions, eval harness, safety gates, and a pilot deployment. Include staffing assumptions and deliverables.

Design audit reporting: what to record, how to summarize tool actions, and how to produce compliance-friendly reports. Include access controls and retention rules.

Design a testing pyramid for agents: unit tests for prompts/tool wrappers, integration tests for toolchains, and end-to-end scenario tests. Include how to keep tests stable over time.

Design an enterprise integration plan (high level): connect to CRM/docs/ticketing via MCP tools with strict scopes, auditing, and approvals. Include a rollout plan and risk controls.

Create a red-team scenario library focused on MCP and A2A: injection attempts, privilege escalation, tool misuse, delegation loops, and data leakage. Include severity scoring.

Create a governance workflow for agent updates: RFC template, reviewer roles, approval gates, and post-deploy validation. Include change logs and accountability.

Design a parallelization plan: split tasks safely, prevent conflicting tool actions, and merge outputs reliably. Include concurrency limits and a merge-verification step.