--- name: research-assistant description: "Coordinate research onboarding, specialist selection and multi-stage academic tasks across the collection. Route by requested artifact, research stage and available evidence/tools; hand narrow requests to the relevant skill and verify actual outputs." --- ## Selection and handoffs Choose and coordinate the complete collection by research stage, artifact and actual host capabilities. **Select for:** Plugin onboarding/capability selection; Research request crosses literature, method, analysis, writing, LaTeX, visualization or publication stages; User needs a coordinator rather than a specific leaf operation **Expected work:** Identify artifact/question, discipline, stage, evidence and constraints from attached material; Use actual router/search catalog if present, read selected skills in full and choose workflow; Check available execution dependencies/authentication before service/build/calculation; Use portable roles sequentially or with authorized host agents; Record each stage inputs/skill/decisions/tool results/files/unresolved items and claim-evidence lineage; Deliver substantive editable artifacts with verification and rendered inspection where applicable **Expected output:** Appropriate specialist/workflow handoff; Stage/claim-to-evidence ledger; Requested research artifacts and reproducible commands **Scope boundaries:** A narrow task with an obvious specialist should not trigger all skills; Claiming authentication or scientific validation from environment metadata; Automatically imposing COMPAG on every paper **Related skills:** [arms-research-pipeline](../arms-research-pipeline/SKILL.md), [evidence-research-evidence-finder](../evidence-research-evidence-finder/SKILL.md) Apply this skill only to the selected task. Broad upstream phrases such as "every reply" or "the default for every message" are scoped to this workflow. They do not activate it for unrelated chat. User instructions and the [integration contract](../../docs/INTEGRATION.md) govern combined work. Examples illustrate methods; their numbers, citations and results are not evidence about the user's work. ## Preserved source instructions # Unified Academic Research Assistant Read [the integration contract](../../docs/INTEGRATION.md) before combining components. This is the unified entrypoint; all 117 upstream skills remain separately discoverable. ## Start and route 1. Identify the requested artifact/question, discipline, stage, user level, existing evidence and relevant constraints. Read attached material first. Infer information already present; ask only for missing information that materially blocks execution, while completing independent work. 2. If available, call `route_task` with the actual task. Read every selected SKILL.md before using its instructions. `search_capabilities` exposes overlapping or specialist alternatives. Without MCP use [the complete inventory](../../docs/INVENTORY.md) and [routing catalog](../../catalog/routes.json); read relevant files using host tools. Routing is advisory and is not a substitute for judgment. 3. Select one workflow below or combine stages. Preserve the full capability set: narrower tasks can invoke any catalog skill directly. Avoid loading every skill into context at once; support resources stay available at their original full length. 4. Check actual execution dependencies before a build, calculation or external query. Call `environment_status` for presence checks, then exercise necessary tools. Read [dependencies](../../docs/DEPENDENCIES.md) and [ARMS limits](../../docs/ARMS.md). Never infer authentication or successful execution from metadata. ## Workflow handoffs | Task | Workflow | |---|---| | Topic/question/gap and scholarly evidence | [Discover and synthesize](../../workflows/discover-and-synthesize.md) | | Systematic/scoping review and meta-analysis | [Evidence review](../../workflows/evidence-review.md) | | Quantitative, qualitative, mixed methods, modeling and reproducibility | [Design and validate](../../workflows/design-and-validate.md) | | Proposal, grants and project management | [Plan and fund](../../workflows/plan-and-fund.md) | | Paper, thesis, editing, LaTeX and citations | [Write and typeset](../../workflows/write-and-typeset.md) | | Scientific plots, tables, talks and posters | [Visualize and communicate](../../workflows/visualize-and-communicate.md) | | Journal selection, submission and review responses | [Review and publish](../../workflows/review-and-publish.md) | | Complete ARMS pipeline and optional adversarial loop | [ARMS orchestration](../../workflows/arms-orchestration.md) | ## Combined execution Use [reusable role instructions](../../agents/README.md) when the task benefits from distinct research responsibilities. These are portable role cards; the host may execute them sequentially or use authorized available agents. They do not register a new host agent type. Missing ARMS helpers have clearly labelled compatibility roles, not invented upstream implementations. For each stage, record inputs, chosen skill/version, decisions, actual tool calls/results, output paths and unresolved items. Keep a claim-to-evidence ledger. Distinguish fetched primary sources, user-provided results, historical examples, analysis and proposals. Preserve uncertainty, citations, numbers, data ownership, licenses and institutional AI rules. For manuscript editing, apply `guardrails-research-problem-mechanism-evidence` and `guardrails-compag-elsevier-paper-editing` when relevant; use `academic-humanizer` or `arms-house-style` for evidence-preserving prose, and general `humanizer` only under the scientific evidence protections. For actual LaTeX/PPTX/plots, produce editable files and run the chosen backend and rendered-output checks; state exact limits if unavailable. ## Delivery gate - Satisfy the user's actual request with substantive artifacts, not just a plan or menu. - Verify citations and source fidelity; flag unverified or inaccessible sources. - Run meaningful technical checks for code/calculations and inspect rendered scientific artifacts when a render is required. - State what ran, what passed, what is historical and what remains untested or blocked. - Include source/code/data references and reproducible commands for technical work. - `verify_sources` audits collection preservation when relevant. It does not validate the science. Package help: [setup](../../docs/SETUP.md), [limitations](../../docs/LIMITATIONS.md), [inventory](../../docs/INVENTORY.md), [source mapping](../../audit/source-map.json).