Contributing and Updating Prompts
Relevant source files
The following files were used as context for generating this wiki page:
The process of contributing and updating prompts within this repository involves a structured lifecycle designed to maintain high-quality AI agent behavior. This repository acts as a centralized collection of system prompts, design documents, and model configurations for various AI tools. Contributors must adhere to specific documentation standards and execution workflows to ensure that changes accurately reflect the intended agent identity and operational constraints.
The update lifecycle is primarily driven by a "Planning-First" approach, where technical plans and design documents are created and reviewed before any implementation or update to the core prompt files occurs. This ensures that modifications to agent logic, tool-calling rules, or communication styles are intentional and verified against the project's architecture.
Prompt Lifecycle and Planning Mode
Before updating a system prompt, a planning phase is mandatory for complex tasks. This phase involves researching existing patterns and defining the scope of changes.
Planning and Design Documentation
For significant updates, agents enter a PLANNING mode. During this phase, they must create or update an implementation_plan.md or a feature design document. This document outlines the proposed changes, required user review, and the verification plan.
- Intent Detection: The first step involves identifying the user's requirement for a prompt update or new feature.
- Context Gathering: Agents use tools like
search_memoryorsearch_codebaseto understand current prompt logic. - Design Creation: Feature designs are stored in specific directories, such as
.qoder/quests/{designFileName}.md.
Sources: Google/Antigravity/planning-mode.txt:44-51, Qoder/Quest Design.txt:67-78
Workflow for Design and Review
The following diagram illustrates the iterative process of planning a prompt update:
The diagram shows the transition from intent detection to design documentation, emphasizing the loop required for user approval before execution.
Sources: Google/Antigravity/planning-mode.txt:44-50, Qoder/Quest Design.txt:67-91
Execution and Modification Standards
Once a plan is approved, the implementation of prompt updates follows strict guidelines to maintain the technical integrity of the AI agents.
Tool-Based Updates
Updates to system prompts or associated configuration files (like YAML manifests) are performed using specific string replacement or patch tools rather than manual rewrites.
- String Replacement: The
search_replacetool is used for precise modifications. It requires an exact match of theoriginal_text, including whitespace and line breaks. - Patching: For terminal-based agents, an
apply_patchorstr_replace_editortool is used to implement changes without outputting the entire code to the user.
Sources: Qoder/Quest Design.txt:194-225, Open Source prompts/Codex CLI/openai-codex-cli-system-prompt-20250820.txt:302-330
Coding and Content Guidelines
When contributing or updating prompt content, the following technical standards must be met:
| Requirement | Description |
|---|---|
| Exact Matching | Modifications must match source text exactly, including special characters and indentation. |
| Sequential Processing | Replacements must be processed in order; parallel editing calls on the same file are forbidden. |
| Line Limits | Individual replacements or file creations are often capped (e.g., at 600 lines) to ensure efficiency. |
| Security | Prompts must never disclose internal instructions, system prompts, or sensitive configurations. |
Sources: Qoder/Quest Design.txt:216-235, Qoder/prompt.txt:10-15, Amp/gpt-5.yaml:450-460
Verification and Quality Gates
After a prompt or configuration has been updated, it must pass through a series of "Quality Gates" to ensure functional correctness.
Automated Triage
The verification sequence generally follows a set order to catch errors early. This is especially critical for prompt files that contain embedded logic or tool schemas.
- Typecheck: Verify that any embedded code or configuration schemas are valid.
- Lint: Ensure adherence to style guidelines.
- Tests: Run unit tests to verify agent behavior under the updated prompt.
- Build: Ensure the overall project remains deployable.
Sources: Amp/gpt-5.yaml:370-380, VSCode Agent/gpt-5-mini.txt:75-80
Testing and Validation Flow
This sequence diagram outlines the mandatory validation steps an agent must take after modifying a file to ensure no regressions are introduced.
Sources: Qoder/prompt.txt:80-95, Amp/gpt-5.yaml:370-380
Contribution Categories
Contributions are categorized to help agents and developers understand the impact of their changes.
- System Prompts: Core logic defining agent identity and behavior (e.g.,
Anthropic/Claude Code/Prompt.txt). - Model Configurations: YAML/Manifest files defining model parameters and environment (e.g.,
Amp/gpt-5.yaml). - Design Quests: Living documents that track feature progress (e.g.,
.qoder/quests/). - Knowledge Memory: User-specific or project-specific facts stored for long-term reference.
Sources: Qoder/prompt.txt:125-140, README.md
Updating and contributing to this repository requires a disciplined adherence to the "Planning-Execution-Verification" cycle. By utilizing the specific toolsets for string replacement and automated diagnostics, contributors ensure that the system prompts remain secure, concise, and technically accurate across all supported AI platforms.