Skip to main content

/tsh-engineer-prompt

info

Not invoked directly by users. To trigger prompt engineering, use /tsh-implement — the Engineering Manager will automatically delegate to the Prompt Engineer.

Agent: Prompt Engineer
File: .github/internal-prompts/tsh-engineer-prompt.prompt.md

Designs, optimizes, audits, or reviews LLM application prompts for quality, security, and consistency.

How It's Triggered​

/tsh-implement <JIRA_ID or task description>

The Engineering Manager identifies LLM prompt tasks in the plan and delegates them to the Prompt Engineer automatically.

What It Does​

1. Context Gathering​

  • Reads the provided prompt(s) or requirements.
  • Identifies the LLM provider, model, and use case.
  • Discovers existing prompt patterns and conventions in the project via tsh-technical-context-discovering.

2. Analysis​

  • Identifies issues — vague instructions, missing output format, injection vulnerabilities, no delimiter separation.
  • Checks against anti-patterns from the tsh-engineering-prompts skill.

3. Optimization or Creation​

  • Applies prompt structure patterns from tsh-engineering-prompts.
  • Adds constraints, output format specification, and security layers.
  • Uses delimiter separation between system instructions and user input.

4. Security Check​

  • Verifies prompt injection defenses — delimiter separation, input sanitization guidance, output validation.
  • Flags any missing security layers.

Skills Loaded​

  • tsh-engineering-prompts — Prompt structure patterns, optimization techniques, security patterns, templates, evaluation approaches, and anti-patterns.
  • tsh-technical-context-discovering — Project conventions and existing prompt patterns.

Output​

Each deliverable includes (sections omitted when not applicable):

  1. Tech Stack — LLM provider, model, temperature, relevant framework.
  2. Prompt Template — Complete prompt with system prompt, context/input sections, and output format specification.
  3. Integration Example — Non-production example showing how to use the prompt (LangChain, OpenAI SDK, Anthropic SDK, etc.). Focuses on integration guidance — full application logic is left to the Software Engineer.
  4. Security Assessment — For audits: vulnerability table with severity, CWE, location, impact, and fix. For creation/optimization: summary of security measures applied (three-layer defense).