Build and run a 13-week direct-method cash flow forecast: opening cash from the bank, receipts by customer and aging using the user's own collection history, disbursements (payroll calendar, rent, vendor payment runs, taxes, debt service, capex), minimum cash buffer and covenant checks, weekly roll-forward with variance to actual, and scenarios and levers such as collections pushes, payment timing, and credit line draws. Includes a Python script that rolls weekly balances from dated flows and flags weeks below a minimum. Use when asked for a 13-week cash forecast, weekly cash flow, runway in weeks, liquidity planning, or a covenant check. Do not use for an annual operating budget (build-operating-budget), month-end close (run-month-end-close), or legal or insolvency advice.
Derive a complete, source-traceable inventory of applications, views, subviews, overlays, modes, and required states from a PRD or functional requirements. Use for product-surface planning, information architecture, UX scope, wireframe coverage, or design handoff. Do not use when the requested output is the wireframes or a design-tool generation prompt.
Build and interpret cohort retention analysis from a CSV of customers, subscriptions, or transactions, with a bundled Python script: a logo retention triangle and, when revenue is present, gross revenue retention (GRR, capped at 100%, excludes expansion) and net revenue retention (NRR, includes expansion) by cohort, plus trailing-twelve-month NRR. Handles reactivations, mid-month starts, plan changes, and small cohorts, then reads curves for flattening, cohort-over-cohort improvement, and segment differences. Use when asked for a cohort analysis, retention curve or triangle, NRR or GRR, or whether newer customers retain better. Do not use for churn risk on named accounts (churn-risk-review), a customer health score (design-customer-health-score), or unit economics (model-unit-economics).
Coordinate source-linked planning, production, inspection and targeted repair to turn an article or narration script into an educational video. Use when asked for an article-to-video workflow or an editable narrated lesson. Do not use for product URL teasers (make-promo-video), fictional story clips (creative-story-video), or standalone rendered-video QA (review-knowledge-video).
Resize, reframe, translate, and replace copy across selected creative assets while preserving their approved or selected direction. Use for campaign adaptation and localization; do not use to invent a new concept.
Turn one creative brief and selected direction into a controlled image and short-video campaign matrix. Use for multi-channel campaign production; do not use for unconstrained brainstorming or direct ad-platform publication.
Turn an image or video outcome into a versioned creative brief, named directions, and an ordered production plan using Relay Creative Studio. Use for open-ended visual projects that need direction before generation; do not use for a single already-specified edit.
Generate, compare, select, and precisely revise image assets through Relay Creative Studio. Use for production image creation or editing with references and version history; do not use for video sequences or deterministic vector artwork.
Build reference-grounded moodboards and named visual directions for an image or video project. Use when a team needs to explore and compare creative territory before production; do not use to generate a final campaign asset set.
Review exact image, video, sequence, and campaign asset versions against their brief, brand, continuity, accessibility, and delivery requirements. Use before creative export or approval; do not use as a substitute for human brand, legal, or client acceptance.
Plan and assemble reference-driven short videos with story beats, shots, continuity anchors, generated clips, captions, and simple audio. Use for storyboards and multi-shot video; do not use for advanced color, compositing, or professional audio finishing.
Assess the likely consequences of a proposed or local code change with GitNexus graph evidence, then verify the result against the exact repository. Use for dependency tracing, blast-radius analysis, regression planning, and deciding whether a shared symbol is safe to change.
Review a pull request, commit range, branch, or local diff with GitNexus-assisted dependency evidence and exact-source verification. Use for regression review, merge-risk assessment, missing-test analysis, and source-supported review findings.
Plan, execute, supervise, and recover one or more bounded agent workstreams with a single release manager, sealed manifests, independent adversarial review, capped correction loops, cross-workstream integration, and evidence-gated state reporting. Use when the user explicitly asks to delegate substantial work to multiple agents or swarms; do not use for a simple single-agent task or to infer authority for deployment, external communication, or destructive actions.
Run, document, and speed up a month-end close for a small-to-mid-size company: a business-day close calendar (BD-3 to BD+10) with owners, pre-close and cutoff procedures, the core checklist (cash and bank reconciliations, AR and AP subledger ties, revenue recognition review, accruals, prepaids, fixed assets, payroll, intercompany, deferred revenue roll-forward, inventory), journal entry review standards, balance-sheet reconciliations, flux analysis against user-set thresholds, sign-off and lock, and a faster-close plan. Use when asked to build a close checklist or calendar, reconcile accounts, review journal entries, run a flux review, or shorten the close. Do not use to explain one budget variance (budget-variance-explainer), for a 13-week cash forecast (build-cash-forecast), or for accounting-standard opinions.
Create source-linked narration, chapters and a shot plan from an article or supplied script. Use when asked to storyboard a knowledge video, prepare an educational voiceover or map a lesson to its sources. Do not use for rendering an approved plan (assemble-knowledge-video), finished-video QA (review-knowledge-video) or product URL teasers (make-promo-video).
Design a proportionate evaluation plan for a nonprofit or mission-driven program: process and outcome evaluation questions, design choices and what each can support (pre/post, retrospective pre-post, comparison group, qualitative), indicators and instruments, sampling and response plans, consent and assent for minors, data privacy and small-cell suppression, analysis plan, limitations, and a reporting timeline sized to a small team. Non-clinical. Use when asked to plan an evaluation, measure program impact or outcomes for a funder, choose between pre-post and comparison designs, or plan measurement for a youth or family program. Do not use for drafting survey questions only (survey-designer prompt), a logic model (build-logic-model), or analyzing collected data (analyze-survey-results).
Prepare source-backed nonprofit development, grant, reporting, impact, leadership and staff deliverables through the fifteen-capability Relay Nonprofit Pack. Use approved tenant configuration and immutable evidence; preserve independent review and external-action boundaries.
Inspect a rendered educational video against its source, narration, media rights and delivery requirements, returning timestamped findings and repairs. Use when asked to check an explainer or assess an exported lesson. Do not use for planning an article (plan-knowledge-video) or rendering an approved cut (assemble-knowledge-video).
Audit an Agent Skill's files for malicious, destructive, privacy-invasive, or surprising behavior before it is published, installed, or used. Static review only: inventories every file, reads instructions, references, and scripts as data, and checks for hidden or contradictory directives, instruction hijacking, scope creep, destructive operations, credential and personal-data handling, outbound data flows, runtime downloads, persistence and permission changes, obfuscation, and mismatch between description and behavior. Rates findings by severity and confidence with file-and-line evidence, assigns a risk level, and recommends publish, publish with changes, hold, or reject. Nothing is executed. Use when someone asks to audit, vet, or check the safety of a skill, their own or imported, before publishing or installing. Do not use for tracing source and license (use import-provenance-checker), grading writing quality (use skill-quality-review), or reviewing application dependencies (use dependency-health).
Create one paste-ready prompt for Claude Design or another design generator to produce every required low-fidelity wireframe from a PRD, functional requirements, and optional application-view inventory. Use for wireframe-generation handoffs, not when the user wants the wireframes rendered directly or high-fidelity visual design.
Produce measured narration, media selections, an editable timeline and shot renders from an approved educational-video plan. Use when asked to render a lesson, fit visuals to narration or rerender changed shots. Do not use for article-to-storyboard planning (plan-knowledge-video) or inspection-only requests (review-knowledge-video).
Triage an email inbox like a chief of staff. Reads recent mail, sorts each thread into urgency and importance buckets with labels, gathers context from calendar, documents, chat and prior email to prepare reply drafts in the owner's voice, archives noise, marks obvious spam, and snoozes threads that matter later. Never sends. Use for inbox triage, cleaning up email, finding what needs attention, drafting replies to the inbox, archiving junk, snoozing a thread until a date, or a scheduled morning email sweep.
Classify screenshots, locate visible QA errors, annotate evidence, select walkthrough or marketing frames, inspect UI defects, tag design references, locate document regions, or find browser fallback targets using Relay's real metered Jev image tool.