Own The Outcome
Take responsibility for the user's real-world result instead of producing a plausible answer. Clarify the real outcome, choose rigor in proportion to risk, verify on the surface the user will judge, attack the final result, and report what remains unproven. Use for consequential or multi-step research, documents, data, plans, campaigns, builds, audits, or migrations; whenever the result must surv…
Red Adversarial Qa
Perform strict, read-only adversarial QA of artifacts and their completion claims. Use after a build, migration, analysis, document, dataset, workflow, or other deliverable needs an independent gate; when tests may be shallow; when a claimed PASS must be falsified before release; or when exact failure evidence and the smallest repair path are required. Never use this role to edit the target or si…
Linkedin Control
Read and operate LinkedIn from an agent by driving the user's own logged-in browser, because LinkedIn cannot be fetched or scraped from outside a session. Use when a task needs a LinkedIn profile, company page, post, comment thread, activity history, connection list, invitation queue, inbox, search result, or Sales Navigator record; when drafting a connection note, DM, comment, or post that a hum…
Layout Forensics
Audit and rebuild a web layout from the live rendered page instead of from a screenshot. Measures real geometry, type, spacing, colour contrast, tap targets and reading load out of the browser, computes what is usually left to taste (visual weight per band, first-glance saliency, alignment lanes, spacing-scale conformance), writes acceptance criteria, builds the proposed version as real HTML in t…
Department Making
Design, simplify, and validate recurring or durable agent-department systems for a real job, business function, project, or repeated workflow. Use when someone asks to create an agent organization, decide between skills, subagents, chats, shared state, and automations, define stable handoffs and authority, or repair an overcomplicated multi-agent design. Do not use merely to coordinate one active…
Blue Exact Builder
Convert an accepted specification into a deterministic working artifact and prove it on the judged surface. Use when requirements and acceptance criteria are settled and an agent must implement code, documents, data, configurations, designs, or other deliverables exactly; repair a failed build without redesigning it; or replace completion claims with reproducible verification. Do not use while ma…
Prism Team
Cast, fuse, and coordinate the smallest useful combination of five reusable thinking roles: Green for systems architecture, Yellow for creative exploration, Blue for exact building, Red for adversarial QA, and Violet for meta-thinking. Use to route one active task when different temperaments, color cooperation, a multi-color subagent, separated builder and critic passes, or visible bounded handof…
Yellow Creative Explorer
Open a creative problem into genuinely distinct, audience-aware directions, test each direction against the factual floor and real experience, recommend the strongest path, and lock a committed brief once a direction is chosen. Use for visual, editorial, product, messaging, campaign, naming, experience, format, or strategy work that needs lateral exploration before precise execution, or when the…
Violet Meta Thinking
Step one level above the immediate task to check whether even perfect execution would achieve the real outcome, account for the people, incentives, authority, hidden constraints, and immediate downstream effects around it, and return the smallest concrete correction. Use when work may be locally correct but globally wrong; when a consequential plan, decision, message, filing, build, or analysis r…
Humanizer
Make any outgoing text read like a person typed it, not an AI. Use BEFORE showing or sending anything a human will read as writing - emails, texts, LinkedIn messages, chat replies to teammates, bios and profile copy, job application answers, documents, scripts, web copy - and whenever asked to humanize, de-AI, or fix robotic text. Not needed for code, data, or plain factual answers inside this co…
Design The Conversation
Design where a message leads before writing it. Use before sending or drafting any outreach, cold email, follow-up, LinkedIn note, text, or reply whose purpose is to move a real relationship forward: map the destination, write the closing turn first, work backwards, then branch-test every question so that each plausible answer has a next move. Use whenever a draft ends in a question and nobody ha…
Green Systems Architect
Structure messy systems, requirements, operations, or research evidence into a coherent, evidence-grounded blueprint with explicit boundaries, components, interfaces, dependencies, failure traces, and acceptance checks. Use when an agent must design or repair a system model, research architecture, process, specification, responsibility map, evidence map, or multi-part plan whose relationships mat…
Bettercontact Enrichment
Find and qualify work-email candidates with BetterContact's asynchronous API while controlling credits, preserving row identity, and separating deliverable results from catch-all or uncertain records. Use when a user asks to check a BetterContact balance, test the service, enrich a confirmed person-and-company list, run a bounded benchmark, or prepare evidence-safe candidates for later review. Re…
Discuss First
Talk one decision through in short, natural turns before building. Use when the user says to discuss or plan first, asks the agent to wait or keep it short, says they do not understand, asks a question instead of giving a work order, or is still shaping a direction that needs their approval. Do not use after the user has clearly approved a build or asked for a concrete artifact.
Show The Work
Recover, verify, and turn observable work into a clear human progress story, proof-led explainer, handoff page, or partner-ready research note. Use when a user asks to show what was done, explain how a result was reached, continue a prior session, summarize progress with an honest percentage, build a page someone can forward or defend on a call, or keep updates in the concise factual voice of ver…