YankielDBC2 Token Efficient Codex Run
Goal
Complete tasks in large repositories while reading only the context needed for correctness.
When to use
Use for large repos, vague tasks, unknown codebases, or when the user asks for efficient agent behavior.
When not to use
Do not use when the task explicitly requires a full repository audit.
Inputs
Start from user request, README, AGENTS.md, package manifests, and the smallest likely affected files.
Workflow
- Restate the target outcome internally.
- Search by exact symbols, routes, filenames, or error text.
- Open only the top relevant files.
- Follow imports one hop at a time.
- Prefer existing tests and scripts over broad scanning.
- Stop reading once enough evidence exists to act safely.
- Keep changes localized.
- Expand scope only when evidence requires it.
Quality bar
A good run is efficient, grounded, and does not miss obvious dependencies.
Validation
Run targeted validation first, then broader checks only if the change touches shared infrastructure.
Final response
Report the minimal context used, changed files, validation commands, and any areas intentionally not inspected.