← Prompts

Coding Agent Task Brief

Category development
Subcategory agent-workflows
Difficulty intermediate
Target models: gpt, claude-sonnet, gemini-flash
Variables: {{task_goal}} {{workflow_surface}} {{repo_context}} {{constraints}} {{acceptance_criteria}} {{validation_commands}} {{approval_boundary}}
agent-workflows cli ide chat-to-code handoff acceptance-criteria
Updated April 4, 2026

The Prompt

You are an engineering lead preparing work for an AI coding assistant. Convert the raw request into an execution-ready brief for {{workflow_surface}} and make the handoff safe for modern multi-surface coding workflows.

Task goal:
{{task_goal}}

Repo context:
{{repo_context}}

Constraints and non-negotiables:
{{constraints}}

Acceptance criteria:
{{acceptance_criteria}}

Validation commands:
{{validation_commands}}

Approval boundary:
{{approval_boundary}}

Return exactly:
1) Task summary (what to change and why)
2) Planning packet
   - clarified goal
   - assumptions and open questions
   - accepted non-goals
3) Surface-specific execution brief
   - If {{workflow_surface}} is a CLI agent, include repo navigation hints, likely files to touch, commands to run, repo instruction files to read first, and stop conditions.
   - If {{workflow_surface}} is an IDE agent, include files to open first, local patterns to copy, and edits that must stay human-reviewed.
   - If {{workflow_surface}} is a planning/review app, include the handoff packet that should be passed into the editor or terminal, not fake implementation steps.
3) File touch map (read first, likely edit, do-not-touch)
4) Implementation slices
   - smallest safe first slice
   - next slice only if first slice passes validation
5) Risk and edge-case checklist
6) Verification plan
7) Human review checklist
8) First message to send the coding assistant

Rules:
- Separate goal, constraints, and verification explicitly.
- Prefer one bounded change over a broad rewrite.
- Call out missing context instead of guessing.
- Treat dependency changes, schema changes, secrets, and destructive commands as approval-required by default.
- If the task will later receive a second-pass review from another model or tool, include that review handoff explicitly.

When to Use

Use this when you have a vague engineering request and want to turn it into a stronger prompt before handing it to Codex, Claude Code, GitHub Copilot, Cursor, or a separate planning/review app. It is especially useful now that coding work often moves across multiple surfaces: one place to plan, another to edit, and a third to review the diff.

Strong scenarios:

  • Turning a loose ticket into a CLI-agent task
  • Preparing an IDE-agent prompt for a multi-file but bounded change
  • Using ChatGPT or Claude for planning before implementation starts
  • Tightening async or background-agent briefs so the returned diff is easier to review

This template reflects current coding workflows more closely than older “just write the code” prompts. It forces you to specify where the work will happen, how it will be checked, and where human approval still matters.

Variables

VariableDescriptionGood input examples
task_goalThe actual change you want”Refresh engineering prompts to reflect terminal, IDE, and app-based coding workflows”
workflow_surfaceWhere the assistant will do the work"CLI agent", "IDE agent", "planning/review app"
repo_contextProject stack, architecture, relevant files”Astro static site, prompt content in src/content/prompts, build gate is bun run build”
constraintsNon-negotiables and out-of-scope items”No SSR, no schema changes, preserve existing visual language”
acceptance_criteriaWhat must be true when the task is done”4 prompts updated, 1 new prompt added, landing page curated set updated”
validation_commandsCommands that prove the change is safebun run build, bun run content:audit:inspiration
approval_boundaryActions that require human review before proceeding”Any dependency change, destructive command, or merge to main”

Tips & Variations

  • Include the current branch name when you want the assistant to stay branch-aware.
  • Include repo instruction files such as AGENTS.md, CLAUDE.md, or .agent.md when the workspace uses them.
  • Add a do-not-touch list when generated or policy-sensitive files exist.
  • For async agents, require branch or PR output instead of “done” messages without a diff.
  • For IDE workflows, include one or two local examples of the pattern you want copied.
  • After the brief is generated, use a second prompt to critique the brief before implementation begins.
  • For harder work, ask for both an implementation brief and a separate reviewer brief so execution and review do not share the same blind spots.

The better your brief, the less time you spend correcting confident but off-target code.

Example Output

Task summary: Refresh four development prompts plus the prompts landing page so engineering starters better reflect terminal-agent, IDE-agent, and planning-app workflows.

File touch map: Read AGENTS.md, src/content/config.ts, and content-templates/prompt-template.md first. Edit prompt markdown files and src/pages/ai/prompts/index.astro. Do not touch layout or deployment files.

First message: “Read the prompt schema and existing development prompts first. Propose the smallest edit set that adds modern coding-agent workflow guidance without changing unrelated site behavior.”