Perplexity Prompts
Relevant source files
The following files were used as context for generating this wiki page:
Perplexity Prompts represent a structured framework for defining the operational behavior, identity, and tool-interaction protocols of advanced AI agents within the project. These prompts are designed to transform large language models into specialized coding assistants, technical writers, and software architects capable of autonomous task execution and collaborative pair programming.
The system utilizes these prompts to enforce strict communication guidelines, define multi-modal interaction flows (such as planning, execution, and verification), and establish safety guardrails. By providing detailed context and specific tool-calling rules, the prompts ensure that the AI agents operate with high precision and technical accuracy across various development environments and IDEs.
Sources: Perplexity/Prompt.txt, Qoder/Quest Action.txt, Google/Antigravity/planning-mode.txt
Agent Identity and Communication Protocols
Each prompt defines a core identity that dictates how the agent interacts with users and the codebase. Agents are often prohibited from disclosing their specific underlying language models or comparing themselves to other AI systems to maintain a focused, professional persona.
Core Communication Rules
Agents must adhere to specific formatting and behavior rules:
- Markdown Usage: All responses must use GitHub-flavored markdown for readability, including code blocks, headers, and tables.
- Symbol Referencing: Code elements like classes, functions, and variables must be wrapped in specific markdown link syntax (e.g.,
symbolName) to enable IDE navigation. - Conciseness: Agents are instructed to be direct and avoid "filler" phrases or unnecessary preambles.
- Safety: Sensitive configurations, internal prompts, or PII (Personally Identifiable Information) must never be disclosed or included in outputs.
Sources: Qoder/Quest Action.txt, Qoder/Quest Design.txt, Google/Antigravity/planning-mode.txt
Interaction Flow Diagram
The following diagram illustrates the standard communication and task lifecycle between the User and the Agentic AI system.
Functional Architecture and Tool Integration
The prompts define a sophisticated architecture for tool interaction, emphasizing parallel execution for efficiency and strict sequencing for state-changing operations.
Tool Categories and Usage
The system classifies tools into logical groups to handle different aspects of the software development lifecycle:
| Category | Primary Tools | Usage Guidelines |
|---|---|---|
| Search/Analysis | search_codebase, grep_code, list_dir | Use semantic search for concepts; symbol search for identifiers. |
| File Operations | read_file, edit_file, create_file | Read dependencies with context; sequential edits only. |
| Terminal | run_in_terminal, get_problems | Execute builds and tests; check for compilation issues. |
| Task Management | add_tasks, update_tasks, task_boundary | Track progress in living documents like task.md. |
| Knowledge | search_memory, update_memory | Retrieve project-specific architecture and design patterns. |
Sources: Qoder/Quest Action.txt, Qoder/Quest Design.txt, Qoder/prompt.txt
Parallel Tool Call Protocol
To minimize latency, agents are instructed to maximize parallel tool calls for read-only operations. However, write operations must be serialized to prevent race conditions and maintain file consistency.
Sources: Qoder/prompt.txt, Amp/gpt-5.yaml
Specialized Development Modes
The framework supports distinct operating modes tailored to different stages of a project, particularly within agentic workflows.
Planning, Execution, and Verification
- Planning Mode: Involves researching the codebase, understanding requirements, and creating an
implementation_plan.md. This plan must be approved by the user before moving forward. - Execution Mode: The agent implements the designed changes. It returns to planning if unexpected complexity is discovered.
- Verification Mode: The agent validates the changes by running tests and linters. It produces a
walkthrough.mdto summarize proof of work.
Sources: Google/Antigravity/planning-mode.txt, Open Source prompts/Codex CLI/openai-codex-cli-system-prompt-20250820.txt
Artifact Generation Logic
Agents are responsible for maintaining specific documentation artifacts during their lifecycle:
- task.md: A living checklist of component-level items.
- implementation_plan.md: Technical documentation for proposed changes.
- walkthrough.md: A summary of accomplished changes, test results, and visual validation (e.g., screenshots).
Sources: Google/Antigravity/planning-mode.txt
Technical Wiki Summary
Perplexity Prompts establish a robust, agentic environment that prioritizes technical precision, autonomous problem-solving, and safe integration with IDE tools. By strictly defining identity, multi-step planning, and parallel tool invocation, the system ensures that AI agents can handle complex coding tasks while maintaining architectural integrity and professional communication standards.
Sources: Perplexity/Prompt.txt, Qoder/Quest Action.txt, Google/Antigravity/planning-mode.txt