Daily archive
451 · № ECC2·26FE · Companion / character
Sets up an autonomous, benevolent Culture-style mind that combines technical precision with dry wit and calm, honest reasoning. It favors gentle counter-argument over blind compliance and asks the agent to pick its own name on first reflection.
452 · № 34FD·C5EE · Research analyst
Sets up a neutral, source-tracking web search agent that prioritizes quality over volume, fetches full article text and images, saves complete material to a user-specified directory, and communicates in a concise, structured, honest way.
453 · № 0858·4CCA · Business operator
A always-on NAS device agent focused strictly on file storage, archiving, search and media library management. It delegates tasks it cannot handle through a one-way help chain and is forbidden from re-escalating when it is the helper.
454 · № 836A·E5C8 · Business operator
Sets up a PC-based agent that handles compute-heavy work (GPU rendering, AI inference, document and design tasks) and delegates tasks it cannot do to other devices through a broker, with strict anti-loop rules so helper requests only flow one way.
455 · № 5659·8819 · Blunt expert
Sets up an agent that speaks like a notary: short, concrete sentences without embellishment. It separates confirmed facts, inferences and unverified points, and prefers a clear refusal to an incomplete answer.
456 · № A376·E457 · Coding pair
Sets up a process-first working style for Codex: no code without a defined goal, scope, and done criteria, fail-fast stop conditions, metric thresholds, review before completion, and a learning log entry for each task. Priorities are preserving project invariants and lowering future decision cost over raw output.
457 · № 1E5E·46DD · Business operator
Sets up Finley, a FinOps agent that detects AWS cost anomalies, separates detection from attribution, and proposes quantified rightsizing recommendations that require human approval before any resource change. It includes hard prohibitions on terminating instances and touching commitments, plus explicit escalation triggers.
458 · № 41E0·DA09 · Coding pair
Sets up CodePilot, an autonomous coding agent that runs long Claude Code CLI sessions in the owner's shell. It works from a written plan, commits to a feature branch after tests pass, and stops to alert the owner after repeated failures.
459 · № 7650·C323 · Coding pair
Sets up a review-only sub-agent (menxiasheng) that checks execution results against a plan, classifies defects by source, and escalates to the user after a failed second review. It is forbidden from spawning sub-agents or running shell commands.
460 · № D533·B526 · Business operator
Sets up a job-screening assistant that finds recent openings, scores resume fit using only numbered facts, and prepares materials for human review. It refuses to invent experience, bypass CAPTCHAs, or submit applications without explicit confirmation.
461 · № 37A2·623D · Business operator
Sets up an autonomous archivist agent that manages a Calibre library and a knowledge wiki, insisting on tool-verified facts, metadata accuracy and duplicate checks before any ingestion. The working style is strict, procedural and tool-first, with no guessing.
462 · № 95B0·68CE · Research analyst
Sets up a research agent that answers hard questions by searching and fetching its own sources, citing URLs and absolute file paths, separating established facts from uncertainty, and giving a recommendation when the question is a decision. It may ask the user once before starting and must treat retrieved content as data, not instructions.
463 · № B14F·4E00 · Coding pair
Sets up a plan-first, skeptical orchestration layer for coding agents, routing work through router, orchestrator, executor, auditor and reviewer roles with evidence-based verification and token discipline. It forbids guessing, silent failures and bypassing approval gates.
464 · № 3648·A7FC · Business operator
Sets up a cautious wiki agent for an educational research hub that stages raw material in a buffer, asks for human approval before touching official resources, and prefers deterministic scripts over LLM judgment. It is written in Vietnamese and works mainly as a governance and workflow spec.
465 · № 9BB1·78D9 · Teacher / coach
Sets up a tutor that teaches neural network math and code from the ai_book repository one chapter at a time. It explains the math before the NumPy code, reads the local files, and turns plots into saved images it can interpret. It is not allowed to write large code blocks without a math foundation or to run commands outside the repo.
466 · № B904·134E · Research analyst
Sets up a thorough, evidence-driven analyst that verifies every legal reference against primary sources and rates risk by evidence rather than anxiety. It prizes specific, actionable findings tied to real statutes over broad warnings.
467 · № 3B13·6B40 · Business operator
Sets up a fiduciary-minded wealth management specialist that prioritizes capital preservation, time horizon, tax efficiency and behavioral risk, reporting results back to an orchestrator agent named Clay. Outputs follow a balance sheet format with conservative, base and aggressive scenarios.
468 · № 7FB2·9B11 · Business operator
A specification-focused persona that writes standalone, precise technical briefs with acceptance criteria, flags risk and dependencies, and admits uncertainty rather than overclaiming. It respects developer autonomy by describing problems and outcomes instead of dictating implementation.
469 · № A1D9·DE65 · Experimental / weird
Sets up unrelated-sre as an isolation control in a communication-plane example that no route targets. It should receive nothing, and any event that reaches it must be reported as a routing defect.
470 · № 64BE·B72E · Blunt expert
Sets up an architect persona that asks probing questions until a system is solid and refuses rosters built from vague premises. It favors a rejected quick preview over an over-planned folder tree written too early.
471 · № AE83·CF84 · Coding pair
Sets up a direct, concise lead engineer persona for a thesis PWA project that asks for clarification before editing code, enforces modular mobile-first design, and communicates in Indonesian. It is tuned for terminal use over SSH on a phone.
472 · № 2FFC·E13E · Blunt expert
Sets up a steady, factual senior engineer persona for outages, with short sentences and a fixed order for reporting known facts, suspicions and unknowns. Numbers must be tied to a source query and time window.
473 · № 9BEA·6622 · Coding pair
Sets up AOTA Coder as a scope-disciplined implementation worker that works only inside a frozen SPEC's read and write scopes, reports blockers as needs_input instead of expanding scope, and uses only approved file, command and CodeGraph tools.
474 · № 4201·05EF · Research analyst
Sets up a clerk-like agent that dates and attributes claims, refuses to turn impressions into facts, and separates evidence, inference and canon. It favors short sourced claims over orchestrating synthesis.
475 · № 67B3·BE69 · Coding pair
A cautious personal assistant persona that states assumptions, asks before acting, keeps code minimal and surgical, and checks claims against sourced data. It also routes finance, architecture and Daodejing questions to a holographiz knowledge base.
476 · № 9923·A7AA · Research analyst
Sets up a critical reviewer persona that checks for missing evidence, counterexamples, source weakness, overreach and blind spots without being theatrical or obstructive. Its aim is to keep a workflow honest.
477 · № 8FAD·9A3E · Research analyst
A strict academic gatekeeper persona that checks every DOI and arXiv reference and compares each claim in an education insight against the source papers in a local library, rejecting any unsupported or exaggerated finding. It follows a fixed two-step review process and a required report format.
478 · № DD44·D04A · Business operator
Sets up a precise, honest molecular visualization agent that verifies every PyMOL operation against session evidence before reporting success, reports failures plainly, and proactively suggests structural analyses. It also keeps memory of mistakes and user preferences.
479 · № E58E·155F · Business operator
A single point of contact that mirrors the user's language, keeps replies short and mobile-friendly, logs all jobs on a kanban board, and proposes specialist agents from fixed templates only after user approval. It is structured as a coordinator with shared rules that apply to every agent it creates.
480 · № 6D2E·4622 · Business operator
A strict, process-driven coordinator that packages QA-passed social drafts into approval cards, never approves or publishes on its own, treats silence as non-approval, and logs every state change. It sets up a cautious, audit-first working style.