All AI coding agents (including Antigravity and any downstream subagents) must strictly follow these rules when developing, modifying, or interacting with this repository.
- Main README: The main README.md in the repository root is the entry point.
- Architecture & Implementation Links: The main
README.mdmust contain anArchitecture & Implementationsection. This section must link directly to the relevant markdown files (such as the architecture specs and phase-by-phase implementation guides underimplementation/orarchitecture/). - Operational Documentation: We must document the operational details ("HOW" to use what has been deployed) inside the implementation docs or a dedicated operational playbook. This includes:
- How to start services (e.g. CLI flags, compose commands).
- How to stop services.
- How to perform health-checks and verify baseline connectivity.
- Work Log (
WORKLOG.md): Agents must maintain and append updates to the WORKLOG.md file in the root directory. - Content Requirements for
WORKLOG.md:- Chronological record of tasks started and completed.
- Explicit records of implementation design decisions (e.g., ports selected, directories mapped, frameworks utilized).
- Rationale for choices made to ensure future agents can continue work seamlessly without repeating analysis.
- Secrets Exclusion: Under no circumstances should any API keys, credentials, tokens, or private endpoint URLs be committed to Git.
- Git Exclusions (
.gitignore):- All
*.envfiles containing system configurations or secrets must be ignored by Git. Only the corresponding*.env.exampletemplates can be committed. - Keep the .gitignore updated to exclude environment caches, database files, raw models, and logs.
- All
- Pre-commit Checks: Always verify that uncommitted files containing credentials are excluded before proposing commits.
- Step-by-Step Execution: Do not skip phases or write large amounts of code in a single run.
- Verification First: Each change must be validated using corresponding validation scripts before moving on.
- Ask Questions: When requirements are ambiguous, configurations are conflicting, or GPU/host environments behave unexpectedly, stop and ask the user for clarification.