chore(agents): add a machine-wide Playwright browser resource budget

Playwright disables normal background throttling, so a hidden 5chan page keeps
doing P2P and rendering work after a check finishes. Agents verifying in
parallel across worktrees stacked whole browser engines on one machine.

Add scripts/pw-session.sh, a wrapper that permits one active Playwright browser
at a time and records who holds it:

- The lock is machine-wide, not per-repository, because the contended resource
  is RAM and CPU. Every worktree and checkout shares one slot.
- Acquisition is an atomic mkdir. Stale locks clear themselves: `open` reclaims
  any slot whose recorded browser is no longer `status: open` in
  `playwright-cli list --all`, so an interrupted workflow cannot strand the
  budget. When that list cannot be read the lock is left alone, so a broken CLI
  never silently disables the budget.
- `open` exits 75 when the slot is busy; `--wait[=SECONDS]` blocks instead.
- `close` always stops the browser, even when the lock was already lost, and
  never releases a slot held by a different session.
- `status` reports the holder and whether its browser is still alive.

Agent policy now runs browser engines and profiler batches sequentially, uses
Chrome/Blink during iteration and the full engine matrix only for final
verification, and never uses `close-all` or `kill-all` while other agents may
own sessions.

Covered by scripts/pw-session.test.js.

docs/agent-playbooks/low-spec-verification.md

@@ -30,7 +30,7 @@ Throttling is **Chromium-only** (CDP). It does not work on `firefox` or `webkit`
 
 ```bash
 # open a Chromium session, throttle it to a mid-tier phone, then verify as usual
-playwright-cli -s=lowspec open https://5chan.localhost --browser=chrome
+./scripts/pw-session.sh open lowspec https://5chan.localhost --browser=chrome
 ./scripts/pw-throttle.sh lowspec mid
 playwright-cli -s=lowspec snapshot
 playwright-cli -s=lowspec screenshot --filename=lowspec-mid.png
@@ -43,7 +43,7 @@ playwright-cli -s=lowspec screenshot --filename=lowspec-mid.png
 
 # reset and finish
 ./scripts/pw-throttle.sh lowspec off
-playwright-cli -s=lowspec close
+./scripts/pw-session.sh close lowspec
 ```
 
 ## Measuring, not guessing
@@ -59,5 +59,6 @@ playwright-cli -s=lowspec eval "() => Math.round(performance.getEntriesByType('n
 ## Caveats
 
 - Chromium-only. Skip on Firefox/WebKit sessions; keep those checks unthrottled.
+- Low-spec emulation is a measurement pass, not a machine-resource control. Hold the machine-wide browser slot for the whole pass and close it immediately afterward.
 - The `low` latency is intentionally aggressive; if requests time out, fall back to `mid`.
 - For render/rerender hotspots after a slow result, use the `profile-browsing` skill (it drives `playwright-cli` + react-scan) on the already-throttled session.

docs/agent-playbooks/skills-and-tools.md

@@ -77,7 +77,19 @@ When using `playwright-cli` for repo UI verification, run the relevant flow in a
 - `firefox` for Gecko
 - `webkit` for Safari/WebKit coverage
 
-Use separate named sessions per engine so results stay isolated. If an engine is intentionally skipped, record why.
+Use separate named sessions per engine so results stay isolated, but run those sessions sequentially. Only one Playwright browser session may be active at a time, machine-wide, because the contended resource is machine RAM and CPU rather than the repository. Open and close sessions through `./scripts/pw-session.sh`; it holds that shared lock so concurrent agents defer and retry browser work instead of saturating the machine.
+
+During iteration, use Chrome/Blink only. Run the full Chrome, Firefox, and WebKit sequence once the change is ready for final verification. Reuse each engine session for desktop and mobile by resizing it, close it in a finally-style cleanup, and only then open the next engine. Do not run profiler batches in parallel, and do not use `close-all` or `kill-all` while other agents may be active.
+
+```bash
+./scripts/pw-session.sh open verify-chrome https://5chan.localhost --browser=chrome
+playwright-cli -s=verify-chrome snapshot
+playwright-cli -s=verify-chrome resize 375 812
+playwright-cli -s=verify-chrome snapshot
+./scripts/pw-session.sh close verify-chrome
+```
+
+When the slot is busy, `open` exits 75; block on `./scripts/pw-session.sh open --wait[=SECONDS] ...` (default 300s) instead of retrying by hand. A lock left behind by an interrupted workflow is reclaimed automatically, because `open` drops any slot whose recorded browser is no longer running. Inspect the holder with `./scripts/pw-session.sh status`, which reports whether that browser is still alive; `release <session>` is a last resort for the rare case where `status` cannot verify the browser state.
 
 ```bash
 npm install -g @playwright/cli@latest