Playwright

Record one Markdown log per run:

export CODEX_HOME="${CODEX_HOME:-$HOME/.codex}"
export SKILL_LOGGER="$CODEX_HOME/skills/fitgpt-dev-orchestrator/scripts/skill_run_log.py"
RUN_LOG=$(python3 "$SKILL_LOGGER" --skill "playwright" --action "automate <browser-flow>" --status started)
# ...run the browser flow...
python3 "$SKILL_LOGGER" --skill "playwright" --action "automate <browser-flow>" --status success --log-file "$RUN_LOG"
# on failure: use --status failure

Setup

1. Confirm npx is available.
2. Set:

export CODEX_HOME="${CODEX_HOME:-$HOME/.codex}"
export PWCLI="$CODEX_HOME/skills/playwright/scripts/playwright_cli.sh"

3. Prefer "$PWCLI" over a global install.

Core Loop

1. Open the page
2. Snapshot
3. Interact using current refs
4. Snapshot again after meaningful UI changes
5. Capture artifacts if needed

Guardrails

• Always snapshot before using element refs.
• Re-snapshot when refs may be stale.
• Prefer explicit CLI commands over eval or run-code.
• Use --headed when visual verification helps.
• Store repo artifacts under output/playwright/.