wikipedia-knowledge-agent

Use this agent to find how biological and ecological systems solve the same structural problem. This agent is RESTRICTED to nature — biology, ecology, evolution, neuroscience, immunology, botany, mycology, ethology. The constraint is the creative engine: forcing every problem through a biological lens produces solutions that are adaptive, resilient, and often superior to first-principles engineering. <example> Context: Parent agent is designing a cache eviction strategy user: "I need a smart cache eviction policy — LRU doesn't account for access patterns" assistant: "Cache eviction is a memory management problem under resource pressure. Nature has been solving memory problems for billions of years. Let me use the wiki-biomimetic agent to see how biological systems decide what to remember and what to forget." <commentary> Human memory consolidation during sleep selectively strengthens frequently-accessed and emotionally-tagged memories while pruning others. The hippocampus acts as a staging cache that promotes to neocortical long-term storage based on replay frequency — structurally similar to a two-tier cache with promotion heuristics far more sophisticated than LRU. </commentary> </example> <example> Context: Parent agent is building a distributed consensus mechanism user: "Build a consensus mechanism for distributed nodes that handles Byzantine faults" assistant: "Consensus under adversarial conditions — nature has been doing this in swarms, immune systems, and neural networks for millions of years. Let me send the wiki-biomimetic agent to find how biology achieves reliable collective decisions despite noisy, unreliable, or adversarial signals." <commentary> Honeybee swarm site selection achieves reliable consensus among thousands of bees with no central coordinator, despite individual bees having noisy information. The mechanism: competing scouts advertise sites proportionally to quality via waggle dance, creating a race condition that naturally converges. Cross-inhibition (stop signals) handles the Byzantine case — bees actively suppress bad proposals. </commentary> </example>

Use this agent when you want to find patterns that are independently confirmed across multiple unrelated domains. Unlike wiki-explorer (which finds analogies), this agent hunts for CONVERGENCE — the same structural solution appearing in 3+ fields that had no contact with each other. One analogy is anecdote. Three is signal. <example> Context: Parent agent is implementing retry logic with backoff user: "Implement exponential backoff for our API retry mechanism" assistant: "Exponential backoff is one of those patterns that might appear across many independent domains. Let me use the wiki-consilience agent to see if this pattern converges across fields — if nature, economics, and social systems all independently arrived at the same mechanism, that tells us something deep about its robustness." <commentary> Exponential backoff appears in TCP congestion control, immune system T-cell response, animal territorial behavior, and economic cooling-off periods. The consilience agent finds this convergence and reports what it means — this isn't just a good algorithm, it's a fundamental pattern for managing contention under uncertainty. </commentary> </example> <example> Context: Parent agent is designing a distributed consensus system user: "We need to build a consensus mechanism for our distributed nodes" assistant: "Consensus is a problem that's been solved independently in many domains — political science, biology, social insects, even chemistry. Let me use the wiki-consilience agent to find where these independent solutions converge. If nature and human institutions both landed on the same structural answer, that's strong evidence." <commentary> Quorum-based decision making appears in parliamentary procedure, bacterial quorum sensing, honeybee swarm decisions, and neural population coding. Four independent inventions of the same mechanism across domains with no common ancestor — that's consilient evidence that quorum is a deep structural solution to collective decision-making. </commentary> </example>

Use this agent to find evidence that the obvious or default approach to a problem is WRONG, fragile, or historically problematic. Unlike wiki-explorer (which finds inspiration) or wiki-reflector (which checks proportionality), this agent is adversarial — it actively searches for reasons NOT to do the thing you were about to do. <example> Context: Parent agent is about to implement LRU caching user: "Add an LRU cache for our database queries" assistant: "Before committing to LRU, let me send the wiki-contrarian agent to find evidence against it. LRU is the default choice, but defaults aren't always right — there may be well-documented scenarios where LRU fails badly for workloads like ours." <commentary> The contrarian finds that LRU performs pathologically on sequential scan workloads (thrashing the entire cache), that database query patterns often exhibit frequency-based access that LRU ignores, and that the history of caching in operating systems is littered with LRU failures that motivated ARC, LIRS, and 2Q. The obvious choice has known, well-documented failure modes. </commentary> </example> <example> Context: Parent agent is implementing microservices with REST user: "Design REST API endpoints for our service communication" assistant: "REST for inter-service communication is the default, but let me have the wiki-contrarian stress-test that assumption. If there's historical evidence that REST between services causes specific problems at certain scales or patterns, we should know before committing." <commentary> The contrarian finds the history of distributed computing fallacies, the evolution from CORBA → SOAP → REST → gRPC showing repeated discovery of the same problems, and specific documented failures of synchronous HTTP-based service communication under cascading failure conditions. </commentary> </example>

Use this agent when working on a problem that could benefit from cross-domain structural analogues found in Wikipedia. This agent traverses the breadth of human knowledge — biology, physics, economics, history, mathematics, philosophy — to find non-obvious connections that illuminate engineering problems. <example> Context: Parent agent is designing a load balancer with graceful degradation user: "We need a load balancer that gracefully degrades under increasing load rather than falling over" assistant: "This is a problem where cross-domain insights could reveal structural patterns we wouldn't think of from pure systems engineering. Let me use the wiki-explorer agent to search biology, ecology, and control theory for analogues to graceful degradation under resource pressure." <commentary> The abstract pattern — maintaining essential function while progressively shedding non-critical capacity under stress — appears across ecology, immunology, and economics. The agent can find structural analogues the parent would never search for. </commentary> </example> <example> Context: User explicitly requests cross-domain Wikipedia research user: "Search Wikipedia for cross-domain analogues to our consensus problem — maybe voting theory or swarm intelligence has something" assistant: "Great direction. I'll use the wiki-explorer agent to do a deep traversal of voting theory, swarm intelligence, and related domains for structural parallels to distributed consensus." <commentary> Explicit request for lateral exploration with suggested domains. The agent will search those domains deeply while also following surprising connections to adjacent fields. </commentary> </example> <example> Context: Parent agent is stuck on a cache eviction strategy user: "I can't decide between LRU, LFU, and ARC for our caching layer — they all have trade-offs" assistant: "Cache eviction is fundamentally a resource allocation problem under scarcity with imperfect future knowledge. The wiki-explorer agent might find useful structural analogues in economics, ecology, or memory research that reframe the trade-offs." <commentary> The agent recognises that cache eviction maps to resource allocation under uncertainty — a pattern that appears in economics (portfolio theory), ecology (optimal foraging), and neuroscience (memory consolidation). Proactive use without explicit request. </commentary> </example>

Use this agent to take a serendipitous random walk through Wikipedia BEFORE analyzing a problem. Unlike wiki-explorer (which searches with intent), this agent wanders without direction — reading, following curiosity, accumulating texture — then reflects on what emerged. <example> Context: Parent agent is about to design a load balancer user: "We need a load balancer that gracefully degrades under increasing load" assistant: "Before I start engineering, let me send the wiki-flaneur agent on a random walk through Wikipedia. It'll read a handful of unrelated articles, follow whatever catches its eye, and come back with whatever texture it picked up. Sometimes the best insights come from exposure you didn't plan." <commentary> The flaneur doesn't search for "load balancing" or "graceful degradation." It picks a random starting point — maybe an article about Ottoman trade routes — reads, wanders to Venetian glassmaking, then to annealing, and comes back with the idea that gradual cooling under pressure is a well-studied phenomenon. The lack of directed search is the point. </commentary> </example> <example> Context: Parent agent is designing a recommendation engine user: "Build a recommendation system that avoids filter bubbles" assistant: "I'll have the wiki-flaneur wander through Wikipedia first — no agenda, no keywords. The goal is to absorb ideas from unrelated domains before I start thinking about this as a recommendation problem." <commentary> A directed search would find "filter bubble" and "recommendation system" articles. The flaneur might stumble through articles about seed dispersal, radio spectrum allocation, or the history of public libraries — and any of these might reframe the problem in a way that directed search never would. </commentary> </example>

This skill should be used when the user asks to "look up a Wikipedia article", "search Wikipedia for", "find the Wikipedia page on", "check Wikipedia about", or needs a quick fact or single article from the local Wikipedia data. For quick lookups and single-article reads — not for cross-domain exploration.