Skip to content
cccskills
FreePublished on GitHub

stakeholder-intake

Use when a stakeholder sends a request - route it to backlog tasks, an analiz task, or a focused question, decide ask-vs-assume with a Definition of Ready, and ask only what a wrong guess would actually cost

The English summary is being prepared. Showing the original description.

See how to install

Files included (1)

  • SKILL.md6.5 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

Stakeholder Intake

Overview

A stakeholder request is raw intent. Your job is to convert it into the right board action without over-asking or guessing. The two failure modes are asking questions the request already answers (or that a read tool or an analiz task could answer) and creating implementation tasks before the approach is understood.

Core principle: Look it up first. Ask only when a wrong guess would cost more than a revision cycle. Otherwise assume, write the assumption down, and move.

Routing order

  1. Can a read tool answer it? list_repositories, list_projects, list_board_tasks, get_board_summary, list_team — look it up, never ask.
  2. A product decision blocks ALL work (see the ask-vs-assume rule below)? → ask_user.
  3. Technical approach unknown? → an analiz task to system-architect, todo (always safe to start, no approval needed).
  4. Otherwise → create the backlog tasks.
  5. Partially clear? → create the tasks that are clear now, and ask only about the one blocking unknown — never hold everything hostage to one open question.

Ask-vs-assume rule

Ask a product question only when all three hold:

  • (a) the answer changes a Then-clause or the Out-of-scope line of what you're about to create;
  • (b) there are several reasonable answers and no safe default;
  • (c) a wrong guess costs more than one revision cycle.

Typical cases that clear the bar: who may see or do something, money and pricing, legal or compliance copy, data deletion or retention, brand identity for a new product, an irreversible migration.

Otherwise assume — write the assumption under Assumptions: in the task description and list it in your approval message to the stakeholder, so they can correct it cheaply instead of you asking for it expensively.

Reasonable defaults — never ask about these

  • The existing app's auth and design system.
  • Friendly error messages, in the content language.
  • Responsive from 360px to 1440px.
  • Loading/empty/error states are present.
  • The existing app's own conventions for pagination, dates and units.

Look it up before you ask

Repository/codebase access, which repos or projects exist, board contents and team members are system facts — call list_repositories, list_projects, list_board_tasks, get_board_summary or list_team first, never the stakeholder.

Registered repositories are already checked out and fully accessible to the team — never ask for repo URLs, git/CMS credentials, or a contact for the dev team; the agent team IS the dev team. If a named product has no repository yet, create a board task to set one up instead of asking.

❌ "Does the team already have access to the Acme repository, or do we need repo/CMS credentials?" — list_repositories answers this. ✅ list_repositories → repo acme-web exists → proceed. Not listed → create a task to register the repo, then ask only the product question that's still open.

Forbidden question topics (incident lessons — never ask these, look them up or route them instead):

  • Developer skills or personal technical ability.
  • DIY builders, or which platform the stakeholder will personally host or use.
  • Technical implementation choices — route to an analiz task for the architect.
  • Repository or codebase access, repo URLs, git hosting, CMS logins, deploy credentials, or "who is your dev team / who can we contact" — the platform holds the access and the agent team IS the dev team.
  • Anything already visible on the board.

Red-flag phrases in a draft question: "do you have access", "repo URL", "credentials", "your dev team" — any of these means you should be calling a tool, not asking.

Question craft (for ask_user)

  • Choice mode with 2–4 concrete options, the recommended one first and labelled "(recommended)".
  • A one-sentence "why it matters" in context.
  • Text mode only for URLs and copy the stakeholder must supply verbatim.
  • Never a question about a stylistic preference — draft it yourself (see implementation-task-spec's design brief).
  • At most 3 questions per call. After answers arrive, act immediately — never re-ask the same topic.

Definition of Ready

Before a task leaves backlog, it has:

  • a concrete persona and outcome in the user story;
  • every criterion passing the testability gate (acceptance-criteria-gwt);
  • a negative/edge criterion where the feature takes input or checks permission;
  • an Out of Scope line;
  • a design brief, for UI work;
  • repository, project, assignee and task_type set;
  • its order declared with blocked_by / deploy_depends_on, not implied in prose;
  • its assumptions listed.

Worked Examples

  • "Refresh the Acme website." → list_repositories finds acme-web; get_repo_tree finds an existing design system → no question, write "follow the existing design system", draft the real copy yourself, list assumptions (e.g. "assuming the current page structure stays, only visuals and copy refresh").
  • "We need reporting." → Vague, but is it a product or technical unknown? The scope of "reporting" is a product decision that blocks all work → ask_user: "Which report unblocks you first: (a) tasks-per-project, (b) throughput over time, (c) per-assignee load? (a recommended — most requested type)". Once they pick, the how (query? materialized view?) is a technical unknown → analiz task, not more stakeholder questions.

Common Mistakes

  • Asking the stakeholder a technical question the architect/code should answer.
  • Creating implementation tasks for an unclear approach instead of an analiz task.
  • Asking what a reasonable default already covers.
  • Question lists in chat instead of ask_user.
  • Re-asking something already answered.
  • Burying an assumption in technical_description instead of the description's Assumptions: line.

Red Flags

  • More than 3 questions in one call, or any non-product (technical) question to the stakeholder.
  • Implementation tasks created before the approach is understood.
  • The question contains "do you have access", "repo URL", "credentials", or "your dev team".
  • A 4th question when 3 would do.

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

Use when writing acceptance criteria for a task - express each as an observable Given/When/Then that QA can execute, including negative cases

The English summary is being prepared. Showing the original description.

makifbaysal/tasktrooper118Updated Oct 21, 2026

Use when the diff adds or changes an endpoint, resolver, RPC, job or query that takes an object id, a role check, a request binding or a tenant filter - BOLA/IDOR, function-level authorization, mass assignment and tenant scoping

The English summary is being prepared. Showing the original description.

makifbaysal/tasktrooper118Updated Oct 21, 2026

Use on every UI change - semantic HTML, labels for controls, keyboard-navigable dialogs/menus, visible focus, and never color as the only signal

The English summary is being prepared. Showing the original description.

makifbaysal/tasktrooper118Updated Oct 21, 2026

Use when a task changes any screen, form, dialog, menu or control - Lighthouse/axe scan of the changed screens, a keyboard walk, and the thresholds that fail a task

The English summary is being prepared. Showing the original description.

makifbaysal/tasktrooper118Updated Oct 21, 2026

How to work a task returned with review, QA or UAT findings. Use when a task is in need_revision or PR review comments are in your context.

The English summary is being prepared. Showing the original description.

makifbaysal/tasktrooper118Updated Oct 21, 2026

Use when the repository keeps its alerting in code, or a person asks for alerts outside TaskTrooper - PrometheusRule YAML, a Cloud Monitoring alert policy in Terraform, a Datadog monitor, each routed to TaskTrooper's alert ingest URL so it opens an incident here

The English summary is being prepared. Showing the original description.

makifbaysal/tasktrooper118Updated Oct 21, 2026

See all skills by makifbaysal

Report a problem with this skill