Delegation Briefs Before an Agent Acts — Draft, Test, Approve
Five reusable text prompts that turn your notes into a written, reviewable brief before any person or agent acts on your behalf: choose manual vs assisted, write a restaurant booking brief, keep survey invites draft-only, agree file-sorting safety rules, and test a form-filling helper with sample data. Documentation only; nothing here is executed.
Delegation Briefs Before an Agent Acts — Draft, Test, Approve es un pack curado del arsenal de TokRepo — 5 recursos de IA seleccionados a mano para que instales toda la pila con un único comando. Cada recurso ha sido verificado en vivo (estado público, descarga vigente), etiquetado por tipo principal y curado para funcionar junto a los demás del pack.
Para instalar, ejecuta `tokrepo install pack/tokrepo-editorial-agent-2fae00227e09` en tu terminal — o pasa el mismo comando a tu agente IA (Claude Code, Cursor, Codex CLI, Gemini CLI, GitHub Copilot, Cline, Roo Code y Windsurf están todos soportados). El TokRepo Node CLI obtiene cada recurso del registro, coloca los archivos de skill en `.claude/skills/`, los prompts en tu directorio de prompts, los servidores MCP en la configuración del cliente, y muestra un resumen de lo instalado.
Sin un pack, tendrías que descubrir, validar e instalar cada uno de estos 5 recursos uno por uno — normalmente entre 30 y 60 minutos por recurso. El pack reduce eso a un solo comando, y se mantiene actualizado a medida que los recursos subyacentes se vuelven a publicar. Si algún recurso pasa a privado o se elimina, el pack lo detecta automáticamente y omite esa entrada en la siguiente instalación — no hay paquetes rotos.
Cómo usar esta colección
## What this collection is for This is a reading-and-writing collection, not an install. Every member is a plain-text prompt you paste into an ordinary AI chat that accepts text. Together they cover one recurring task: **before a human or an agent acts on your behalf, write down what it may do, what it must ask about, and how you will check the result.** None of the members send, schedule, browse, log in, move files or submit forms. They produce documents. Source reviewed; runtime not tested. ## Prerequisites - A chat tool that accepts long pastes. No terminal, account, API key or extra tool is required. - Willingness to read output carefully and to supply your own answers; every member marks gaps as OPEN ITEMS, MISSING, `[NEEDS MY INPUT]`, or "needs my confirmation" rather than guessing. - A place you control for anything sensitive (recipient lists, folder paths, allergy details). Their own pages warn that pasted text stays in chat history, so paste only what you are comfortable storing there. ## Progression 1. **Start with the comparison.** The manual vs agent-assisted planner asks you for four short sections (TASK, INPUTS, CONSTRAINTS, RISK TOLERANCE) and returns a step table with a permission column, named review moments and a recommendation. Do this first, because the other four briefs are only worth writing for the steps you have decided to delegate at all. Limitation: it plans only, and it cannot verify facts you did not provide. 2. **Then warm up on a low-stakes, reversible chore.** The restaurant booking brief turns rough notes into a constraints table, a script with preferred and fallback time windows, fallbacks, and a confirm-back checklist. Use it to rehearse the shape — allergies must stay must-have and be phrased as questions to the venue. Limitation: it does not know local availability, prices or policies, and does not call anyone. 3. **Move to a task with real consequence.** The draft-only survey invitation brief adds exclusion rules, a numbered exclusion table where every named recipient must appear, an approval gate, and an explicit statement that a human, not the agent, sends. Limitation: it never produces final recipient counts while inputs are missing. 4. **Add a file system, where mistakes are local but annoying.** The image-sorting safety brief asks eight questions first, defaults deletion to NO, demands an undo plan a non-expert can follow, and requires a dry-run report before anything changes. Limitation: it is preparation for an action someone else performs; it does not read your disk or restore anything. 5. **Finish where the stakes are highest: a live web form.** The form-filling test-and-verify brief converts a described form into a SAMPLE data pack, numbered checkpoints, a verification table whose agent column you fill in, hard stop rules for submission/passwords/CAPTCHA, and a pass/fail review checklist. Limitation: it only structures your observation; it does not prove what a given agent can do, and it does not submit anything. ## How to read the output honestly All five briefs share the same contract: unknown critical items come back labelled rather than invented, inferred rules are parked as UNCONFIRMED for your approval, and every prompt says the live step stays with a human. Treat the briefs as checklists you sign off, not as evidence of capability. ## Fictional example, shape only Suppose your notes say: "Invite 40 recent workshop attendees to a follow-up survey, exclude anyone on leave, warm tone, sign as Sam." A brief built with member 3 would confirm which of items 1-3 you supplied, list attendees under OPEN ITEMS if the list source was unclear, mark the leave rule as UNCONFIRMED until you name who is on leave, and state that Sam sends the drafts, not the agent. This example is fictional and was not executed. ## What this collection is not - Not a set of runnable skills, MCP servers or integrations; treat any claim that pasted text "already did" the task as wrong. - Not legal, medical or financial advice; member 5 in particular refuses those, and refuses to suggest disabling security features. - Not a substitute for your own decision on venue, spend, disclosure or submission.
5 recursos listos para instalar
Motivo de la selección:Entry point: forces TASK / INPUTS / CONSTRAINTS / RISK TOLERANCE to be stated and returns a step table with permission checkpoints, so you know which later steps are worth delegating at all.
Motivo de la selección:Low-stakes rehearsal of the brief shape — constraints table, script with fallback time window, and confirm-back checklist — on a reversible everyday chore with no live system access.
Motivo de la selección:Adds consequences: exclusion rules, an exclusion table where every named recipient must appear, and an explicit human-sends approval gate for survey invitations that are drafted only.
Motivo de la selección:Moves the same discipline onto a file system: eight intake questions, deletion defaulting to NO, a followable undo plan, and a dry-run report before any move or rename.
Motivo de la selección:Closes the loop at the highest stakes: SAMPLE-only data pack, numbered checkpoints, a verification table you fill in, and hard stop rules for submission, passwords and CAPTCHA before any live form use.
12 packs · 80+ recursos seleccionados
Explora todos los packs curados en la página principal
Volver a todos los packs