Home › Prompts › Claude › Agentic coding task brief for Claude Code

Agentic coding task brief for Claude Code

A task brief for Claude Code that defines the goal, constraints, verification and stopping point so the agent works autonomously but safely.

ClaudeAgents & automationRuleStack original

Written for Claude Opus 5.5 (Claude Code) · Updated 2026-10-16

Note: Review and test before use. Provided as is, without warranty. See Terms.

Goal: [outcome, e.g. add CSV export to the reports page].
Context: [relevant files/dirs, related issue, design notes].
Constraints: keep changes within [dirs]; follow existing patterns in [example file]; no new dependencies without asking; don't change public APIs.
Plan first: explore the code, then write a short plan and wait for my OK before editing.
Done means: [tests to add], `[test command]` and `[lint command]` pass, and you report what you verified versus what you assumed.
If blocked: stop and ask instead of guessing.

Replace everything in [square brackets] with your own details.

Settings & parameters

• Claude Code with claude-opus-5-5 for complex work, claude-sonnet-5-5 for routine tasks
• Pair with a CLAUDE.md (see RuleStack's CLAUDE.md rules) for project conventions

Variations

• Bug fix: "reproduce with a failing test first".
• Refactor: "no behavior change; keep the diff under [n] lines".

Tips

• A clear definition of done is the biggest lever for agent quality.