start-task

name	start-task
description	Guides the agent on how to start and execute a development task in the Neural-Browser Runtime (dragon-head) project following strict repository rules. Use this when the user asks you to start a task, pick up the next task, or proceed with development.

Start Task Skill

This skill defines the standard operating procedure for starting and completing a development task in the dragon-head (Neural-Browser Runtime) repository. It ensures that planning documents, coding conventions, architectural specifications, and PR rules are strictly followed from start to finish.

When to use this skill

• When the user asks you to "start a task", "pick up the next task", "implement the next phase", or "continue with development".
• When you are tasked with selecting the next valid task from the PLAN.md to work on.

Strategy: Explore -> Plan -> Implement -> Commit

Follow these steps sequentially to execute a task:

1. Explore & Plan

Do not jump straight to coding. Build the necessary context first:

1. Read the Rules: Review docs/PROMPT.md carefully. This document contains absolute rules for branching, committing, testing, error handling, and coding standards. You MUST adhere to them.
2. Review Specifications: Check both docs/PLAN.md and docs/SPEC.md. Understand the current project phase and PR dependencies from PLAN.md.
3. Select the Task: Identify the next logical task/PR to implement that has not been completed yet (e.g., PR-01), based on docs/PLAN.md.
4. Acquire Code Context: Explore the existing codebase (using file viewing, directory listing, or search tools) to fully understand the areas your selected task will modify.
5. Ask/Formulate: If the task scope is ambiguous or modifies multiple core files, create a brief implementation plan and ask the user for verification.

2. Implement & Verify

Ensure you have a reliable way to verify your work:

1. Create Branch: Check out a new branch from the latest main. The branch name must follow the feature/PR-XX-description convention as defined in docs/PROMPT.md.
2. Test-Driven Development (TDD): Perform strict TDD (Red -> Green -> Refactor) as requested. Always start by writing a failing test based on the test tasks in docs/PLAN.md.
3. Write and Execute Code: Write your implementation code in small, incremental steps.
4. Local Quality Gates: Ensure your code passes all mandatory quality checks before concluding.
   • Run tests: cargo test --workspace (or pytest for Python components, or just test-all if available).
   • Run linters: cargo fmt --all -- --check and cargo clippy --workspace -- -D warnings.

3. Commit, PR & Document Update

1. Update Docs: Mark the completed task with [x] in docs/PLAN.md. Update docs/SPEC.md or README.md if necessary.
2. Commit: Ensure any new files are added with git add. Use Conventional Commits (<type>(<scope>): <summary>) for your commit messages.
3. Push: Push your created branch to the remote repository (git push origin feature/PR-XX-...).
4. Create PR: Open a Pull Request for your branch using the GitHub CLI (gh pr create). Ensure the PR title is [PR-XX] 実装の概要 and the description includes all mandatory sections (Exit Criteria status, test results).
5. Monitor CI: Ensure CI passes and fix any issues automatically.