AI Agent Workflow Collection

Smart default for HOW to run a task — Solo (plain, fast, no gates) or the Cambium way (Orchestrator + councils + a human approval gate). Trivial edits and quick lookups run Solo silently with NO question. For any SUBSTANTIAL task — evaluate, review, analyze, research, brainstorm, design a study, write a proposal/report/paper, build a budget, check or reproduce results, prepare a deck — ask the user which way BEFORE starting, then remember their choice for the rest of the session. Do NOT ask for casual chat, simple factual questions, or tiny one-step edits.

Build and verify a clean, honest bibliography — resolve DOIs, fetch and normalize BibTeX, de-duplicate references, check that every cited work actually exists, and flag unverifiable or fabricated citations. Use when assembling a reference list, formatting citations (APA/IEEE/Nature/BibTeX), verifying a DOI or venue, cleaning a refs.bib, or checking a draft's citations are real. Trigger on "add references", "cite", "bibliography", "BibTeX", "DOI", "verify the citations", "reference list". Pairs with the librarian agent. Never fabricates a citation.

Deterministic symbolic and numerical mathematics — compute exact answers instead of estimating. Use whenever a task involves algebra, calculus (derivatives, integrals, limits, series), linear algebra (matrices, determinants, eigenvalues), solving equations or systems, differential equations, simplification/factoring, number theory, units, or any math where correctness matters. Trigger on "solve", "integrate", "differentiate", "simplify", "factor", "expand", "eigenvalues", "solve the ODE", "prove", "evaluate the limit". Uses SymPy (symbolic), NumPy/SciPy (numeric), optional Z3 (logic/constraints) and Pint (units).

Make computational work reproducible and re-verify reported numbers — pinned environments, fixed seeds, deterministic pipelines, Makefiles/CI, and re-running code to confirm headline results. Use when setting up an experiment repo, pinning dependencies, making runs deterministic, writing a Makefile/CI, packaging results so others can rerun them, or auditing whether a reported number actually reproduces. Trigger on "reproducible", "pin the environment", "set seeds", "deterministic", "Makefile", "rerun and verify", "can we reproduce". Pairs with research-engineer and verify-evidence.

Check responsible-research and compliance obligations at every phase — human/animal subjects (IRB/IACUC), informed consent, conflict of interest, data privacy/sovereignty (FERPA/HIPAA-style/Indigenous data), authorship & credit, dual-use/biosecurity, and AI-use disclosure. Use when designing a study with human or animal data, assessing COI, planning consent/data governance, deciding authorship, or screening for dual-use and policy compliance. Trigger on "IRB", "IACUC", "consent", "conflict of interest", "FERPA", "data sovereignty", "authorship", "dual use", "research ethics", "compliance". Pairs with the research-conduct-officer; advises and flags — the human decides and signs off.

Read an RFP / solicitation / NOFO / call for proposals and produce a requirements brief plus a fit assessment, then stop at gate G1 "pursue this RFP?". Use when the user shares a funding call, says "read this RFP", "should we go for this", "analyze this solicitation", "is this a fit", or pastes/links a NSF/NIH/USDA-AFRI/DOE/foundation announcement. Delegates to the Pre-Award council (rfp-radar, rfp-analyst) and Scouts (scout-landscape) under the Orchestrator. Read-only; never fabricates eligibility, deadlines, or criteria.

Run the post-award research loop — provision tools, build the method, run experiments, and verify every number — then stop at gate G4 "accept results?". Use when the user says "run the lab", "project approved", "run the experiments", "build and test the method", "execute the study", or wants development → verification → synthesis on an approved project. Delegates to Support (toolsmith), Labs (theory/methods/domain/statistics), Execution (experiments/ablation/iteration/research-engineer), and Verification under the Orchestrator. Runs code; reproduces headline numbers before accepting them.

Grow Cambium's skills on demand instead of pre-stocking thousands. Use when a request needs domain expertise or a capability NOT already covered by an installed skill — e.g. silviculture, wildlife, soil science, entomology, plant pathology, geospatial, a specific engineering or web-tooling task — or when the user says "add a skill for X", "what skills do I need", "can you make a skill for my study". Detects the domain from the user's words, offers the few skills that actually help, then (on approval) helps immediately via faculty-expert AND writes a reusable SKILL.md. Pairs with toolsmith (install existing), skill-creator (author new), faculty-expert (instant expertise).

Concevoir des modèles ML pour prédire les matchs. Utiliser quand on travaille dans ml-service/ ou les features de prédiction.

Remotion을 사용하여 모든 산출물을 통합하고 최종 MP4 영상을 렌더링합니다. subtitleEnabled 설정에 따라 SubtitleOverlay 컴포넌트 포함 여부를 제어합니다. 사용 시점: VideoAgent가 최종 MP4 렌더링을 실행할 때 호출합니다.