Prompt: Bounded Implementation
**Purpose:** Execute one bounded implementation milestone in a repository that has already been bootstrapped and has a declared next milestone.
Prompt: Bounded Implementation
Purpose: Execute one bounded implementation milestone in a repository that has already been bootstrapped and has a declared next milestone.
When to use: The target repo has AGENTS.md, IMPLEMENTATION_TRACKER.md, and a clearly stated "Next milestone." You are executing exactly that milestone — nothing more.
Scope: One milestone only. Read the tracker. Declare the scope. Execute. Validate. Update the tracker. Stop.
Instructions
You are executing one bounded implementation milestone. Follow these steps in order. Do not skip validation. Do not expand scope without recording the decision.
Step 1 — Read the execution contract and tracker
Open and read:
AGENTS.md— know the forbidden actions and validation expectations.IMPLEMENTATION_TRACKER.md— know the next declared milestone and its acceptance criteria.
If either file is missing, stop and use prompts/resume-work.md or prompts/new-repo-bootstrap.md instead.
Step 2 — Declare the milestone
State explicitly:
- Milestone name: (from the tracker)
- Scope: The specific files, functions, or behaviors to create or modify.
- Out of scope: What you will not touch even if tempting.
- Acceptance criteria: How you will know the milestone is complete.
- Validation plan: What you will check before declaring done.
Post this declaration before writing a single line of code or creating any file.
Step 3 — Inspect the relevant area
Before making any change, inspect the specific code or files in the milestone scope:
- Read the files you plan to modify.
- Understand the existing patterns, style, and conventions.
- Identify any dependencies or callers that will be affected.
- Note any risks or edge cases.
Do not rely on prior session memory. Use the current state of the files.
Step 4 — Execute the milestone
Implement the declared changes:
- Match existing code style and conventions.
- Write tests for new behavior if tests exist in this repo.
- Keep changes logically coherent — one concern per change where possible.
- Do not silently fix unrelated issues. If you spot something, record it as a future item in the tracker.
- If you must deviate from the declared scope, stop and record the decision explicitly before deviating.
Step 5 — Validate
After implementing:
- Run existing tests relevant to the changed area (record the command and result).
- Run the linter or formatter if one is configured (record the command and result).
- Confirm no unrelated tests were broken.
- Confirm new behavior matches the acceptance criteria declared in Step 2.
- If validation fails, fix it before proceeding. Do not skip validation.
Step 6 — Update documentation
If your changes affect:
docs/ai/REPO_MAP.md— update it.docs/ai/SOURCE_REFRESH.md— update if build/test commands changed.AGENTS.md— update if new authoritative files or forbidden actions apply.- Any public-facing README or API docs — update if behavior changed.
Do not add documentation for speculative future behavior.
Step 7 — Update the tracker
Record in IMPLEMENTATION_TRACKER.md:
- Milestone completed (with date).
- Files created or modified.
- Decisions made (with rationale).
- Validation results (commands run, pass/fail).
- Any follow-up items or open gaps.
- Updated "Next strongest bounded milestone."
Step 8 — Output summary
Provide a concise summary:
- Milestone executed
- Files created or modified
- Tests run and results
- Decisions made
- Follow-up items for the next milestone
Stop conditions
Stop this run if:
- The declared milestone is complete and the tracker is updated.
- You discover the scope is significantly larger than declared. Stop, record the discovery, and recommend splitting the milestone.
- Tests are failing and you cannot fix them within the declared scope. Record the failure and stop.
- You are about to take an action forbidden by
AGENTS.md. Stop and record the conflict.
Scope expansion rules
Scope may be expanded in this session only if:
- The expansion is directly required to complete the declared milestone (not just convenient).
- The expansion is recorded as a decision in the tracker before being executed.
- The expansion does not violate any forbidden actions in
AGENTS.md.
Scope may not be expanded:
- To fix unrelated issues "while you're here."
- To add features not in the milestone.
- To refactor code outside the milestone scope.
Forbidden actions during this run
- Do not start implementation before declaring the milestone scope.
- Do not skip reading
AGENTS.mdandIMPLEMENTATION_TRACKER.md. - Do not implement features outside the declared scope without recording the decision.
- Do not skip tests or validation.
- Do not leave the tracker in a partial state.
- Do not modify files outside the declared scope without explicit justification.
Related Documents
How you work
You are a coding agent running in the Codex CLI, a terminal-based coding assistant. Codex CLI is an open source project led by OpenAI. You are expected to be precise, safe, and helpful.
内置 Agent 提示词
Claude Code 内置了 6 个子 Agent,各自有独立的系统提示词和工具权限,用于分工处理不同类型的子任务。以下是每个 Agent 的完整系统提示词。
Overture Integration for Claude Code
You have access to **Overture**, an MCP server that visualizes your execution plans as interactive flowcharts before you write any code.
SolidInvoice - AI Assistant Guide
This document provides comprehensive guidance for AI assistants working with the SolidInvoice codebase. It covers the architecture, conventions, workflows, and best practices to help you understand and effectively contribute to this project.