lz-create-agentsmd

Workflow

AGENTS.md

Phase 1: Deep Semantic & AST/LSP Scan

1. Tech Stack Discovery: Read package manager files (

   package.json

   ,

   .csproj

   ,

   pom.xml

   , etc.) to identify the core languages and test frameworks.
2. Semantic / AST Search: Do not rely on simple regex grep. Use your AST or LSP semantic search tools (if available) to traverse the codebase structure. If not available, do a deep file-read traversal.
3. Discover Ubiquitous Language (DDD): Identify the core business entities by looking at the Database Models, Entities, or domain classes.
4. Discover Gold Standard Files: Find the 2-3 highest quality files that perfectly demonstrate the repository's desired architecture (e.g. a perfect Controller, a perfect Service class). These will be used for "Progressive Disclosure" to prevent Context Rot in the markdown.

Phase 2: User Interview (Chained

ask_question

)

ask_question

1. "I have detected the following core Domain entities:

   [List]

   . Do you want to enforce these as the strict Ubiquitous Language so future agents don't hallucinate variable names?" (Allow them to add/remove terms).
2. "I found

   [File 1]

   and

   [File 2]

   as excellent representations of your architecture. Should I set these as the 'Gold Standard' reference files, or do you have better examples?"
3. "What is the exact CLI command required to run the automated tests with coverage for this project?"
4. "Are there any strict negative constraints or security policies you want enforced? (e.g. 'Never use raw SQL', 'Always use AWS Secrets Manager')."
5. "What are your Git and Workflow conventions? (e.g., branch naming like

   feature/JIRA-123

   , rules against force pushing, commit formats)."

Phase 3: Template Generation

1. Read the

   TEMPLATE_AGENTS.md

   file located in your skill directory to understand the exact structure required.
2. Inject the user's validated answers (DDD terms, Gold Standard file links, Test commands, Security rules) into the template structure.
3. Write the final

   AGENTS.md

   to the root of the user's repository.
4. Print a success message confirming the full-lifecycle engineering guardrails have been applied.