Codex workspace guidance

QA automation resources

Use repository skills from .agents/skills when the request matches their description or when the user invokes a skill with $skill-name.
Read only the relevant rule files before implementing or reviewing automation code:
- .agents/rules/automation_rules.md for shared automation and test-data conventions.
- .agents/rules/locator_strategy.md for locator selection.
- .agents/rules/playwright_rules.md for Playwright work.
- .agents/rules/selenium_rules.md for Selenium work.
- .agents/rules/appium_rules.md for Appium work.
- .agents/rules/delivery_checklist.md before completing automation work.
Invoke workflow skills with $ and kebab-case, for example $generate-locator or $analyze-flaky-tests.

Interpret tool names retained in migrated instructions by intent and use the callable Codex equivalent in the current session.
Map view_file to a read-only filesystem command, write_to_file/replace_file_content to apply_patch, and run_command/command_status to shell execution and its wait mechanism.
Map read_url_content to the available web or in-app browser capability.
Map conceptual browser_* steps to the installed in-app browser skill and its current callable tools; do not invent unavailable tool calls.

Communicate and report results in concise, clear Vietnamese.
Preserve the current local code state. Do not run state-changing Git commands such as git pull, git checkout, git merge, git rebase, or git reset unless the user explicitly requests them. Read-only Git inspection is allowed.
Prefer stable semantic locators and smart waits; do not introduce fixed sleeps unless the user explicitly requires them.
Keep test data unique, traceable, deterministic when seeded, and free of real personal information.
Validate generated automation with the narrowest relevant test, lint, or compile command available in the target project.