Task mode

Paste any task — a Jira ticket, Slack thread, PR feedback, or half-formed idea — and get a structured, expert-grade prompt your AI coding agent can act on without clarifying questions.

What it does

AI agents do their best work when the starting context is precise. Task mode bridges the gap between "raw ticket" and "agent-ready prompt" — it reads your task input, detects the domain and tech stack automatically, and restructures it into a format that gives an AI coding agent everything it needs: context, acceptance criteria, constraints, and explicit out-of-scope boundaries.

The result is a TASK.md you paste directly into Claude Code, Cursor, or any other agent — or a SPEC.md if you need a full engineering specification for planning and estimation.

The 3-step wizard

1

Paste your task

Drop in anything: a Jira ticket description and its comments, a Slack thread, a PR review, a bug report with reproduction steps, or a plain verbal description. The minimum is 20 characters — there is no maximum.

You can optionally expand "Add tech stack context" to paste a package.json, requirements.txt, or a typed list of technologies. If you skip it, MDPilot auto-detects the stack from keywords in your task text (React, FastAPI, AWS, etc.) and shows the detected tags below the input field.

2

Configure output

Tell MDPilot how you will use the output. This step controls the execution mode, experience level, and two optional toggles. MDPilot also shows the detected domain (Frontend, Backend, AWS, DevOps, etc.) so you can confirm it read your input correctly.

See Execution modes and Other options below.

3

Review & generate

A summary shows your task input preview, detected domain, output mode, experience level, tech stack, and which files will be generated. Choose your AI model provider if more than one is configured. Click Generate — each file takes about 5–15 seconds.

Execution modes

The execution mode is the most important choice — it controls the depth and structure of the output. Pick the one that matches who (or what) will use it.

Guide

Developer guide

Full TASK.md with all sections — context, rationale, watch-outs, acceptance criteria. Best when a human developer will read and act on this.

AI Exec

AI execution

Prescriptive TASK.md written for an AI agent to execute directly — exact file paths, function signatures, command sequences. Use this when you're pasting directly into Claude Code, Cursor, or Copilot.

Context

Context drop

Compact output: task + requirements + acceptance criteria only. Use this to paste into any chat window to ground a conversation without generating a full file.

AI Exec is the mode that benefits most from the other options — it unlocks the verification pass toggle and produces the tightest agent-ready output.

Other options

Experience level

Experienced

Experienced — terse

Assumes the developer knows the language and framework. Focuses on the task specifics, not the basics.

New

New to stack — explain why

Adds explanations of non-obvious decisions and patterns specific to the detected stack.

Verification pass

Only available in AI Exec mode. When enabled, MDPilot adds a self-check section to the output so the agent validates its own work before finishing — checking that every acceptance criterion is met, every referenced file exists, and every command runs. Useful for longer tasks where subtle errors are easy to miss.

Show alternative approaches

When enabled, the output includes 2–3 domain-appropriate implementation alternatives with trade-offs and a recommendation. Useful when you want the agent to reason about options rather than execute a single prescribed approach.

Output files

TASK.mdrecommended

The primary output. Contains structured task context, acceptance criteria, constraints, and an agent prompt block ready to paste into your AI tool. Save it at the project root or paste the Agent Prompt block directly into your coding agent.

SPEC.md

Engineering specification — user story, functional requirements, technical approach, and a testing plan. Use alongside TASK.md for planning, estimation, and review before implementation starts.

You can generate both at once — MDPilot makes one API call per file.

Using the output

The output view has three actions: copy the markdown, download the .md file, or download a .zip if you generated multiple files.

The fastest workflow: copy the Agent Prompt block from the output (it appears at the bottom of the TASK.md, clearly labelled) and paste it directly into Claude Code, Cursor, Copilot, or Windsurf as your first message. It gives the agent the task context, the acceptance criteria, and explicit boundaries in one paste — no back-and-forth.

Alternatively, save TASK.md at the project root and reference it in your agent: "Read TASK.md and implement it."

Using via MCP

With the MDPilot MCP server configured, you can generate TASK.md without leaving your IDE: "Use mdpilot to generate a task file for this ticket: [paste ticket]." The MCP tool also detects the stack from your real repo.

Worked example

Input — a rough Slack message:

hey can someone add rate limiting to the /api/export endpoint it's getting hammered, causing timeouts for other users probably 10 req/min per user should be fine we're on Next.js + Upstash Redis already

Settings: AI Exec · Experienced · Verification pass on

Output: a TASK.md with the goal ("implement rate limiting on /api/export"), the detected domain (Backend), constraints (10 req/min per authenticated user, use existing Upstash Redis), acceptance criteria (429 response when exceeded, existing users unaffected, latency <5ms overhead), an implementation plan referencing the actual Next.js middleware pattern, and a verification pass checklist the agent runs before marking done.

The Agent Prompt block at the bottom is a single paste into Claude Code that starts the implementation with zero clarifying questions.