pi-browser-use 0.9.5 → 0.9.6

package/README.md

@@ -159,7 +159,7 @@ On startup the default persistent profile is checked for accessibility. A root-o
 
 `npm run bench` times the tool stack headless over 5 iterations (fixture setup excluded, matching vercel-labs/agent-browser's scenario set). Baseline on Apple Silicon: navigate ~1ms, snapshot ~3ms, screenshot ~37ms, evaluate ~206ms, full agent-loop cycle ~422ms. Re-run on your hardware before quoting numbers.
 
-`npm run perf:audit` reports cold-import memory/time plus published package and dependency footprint; `npm test` enforces the stable regression budgets. See [Performance audit and budgets](docs/performance.md) for the M0 measurements, thresholds, and reproducible local/CI commands.
+`npm run perf:audit` reports cold-import memory/time plus published package and dependency footprint; Code Foundry's dedicated performance job enforces the stable regression budgets with `npm run perf:check`. See [Performance audit and budgets](docs/performance.md) for the M0 measurements, thresholds, and reproducible local/CI commands.
 
 ## Bundled skills
 

package/docs/performance.md

@@ -5,7 +5,7 @@ This project measures performance in two layers:
 - `npm run perf:audit` measures cold plugin import time and memory, the published package, and the production dependency closure. It is fast, emits JSON, and does not launch Chrome.
 - `npm run bench -- --iterations 10 --json` launches the complete headless Chrome + MCP stack and measures startup, process-tree RSS, and representative tool calls against a fixed local fixture.
 
-`npm test` runs `perf:check` after the unit suite, so the stable artifact, dependency, and normalized cold-import budgets run in CI. The audit reports raw time but compares it with a same-run `typebox` import to account for runner CPU variance. Browser timing and process-tree memory are reported rather than enforced because shared-runner and Chrome variance would make a wall-clock budget flaky.
+Code Foundry runs `npm test` and `npm run perf:check` in separate unit and performance jobs, so the stable artifact, dependency, and normalized cold-import budgets remain required in CI without coupling them to the unit command. The audit reports raw time but compares it with a same-run `typebox` import to account for runner CPU variance. Browser timing and process-tree memory are reported rather than enforced because shared-runner and Chrome variance would make a wall-clock budget flaky.
 
 ## M0 baseline and result
 
@@ -33,7 +33,7 @@ Budgets live in [`performance-budgets.json`](../performance-budgets.json). They
 
 | Metric                            |        Budget |
 | --------------------------------- | ------------: |
-| Cold import / `typebox` p50 ratio |           2.0 |
+| Cold import / `typebox` p50 ratio |           3.0 |
 | Cold import maximum RSS delta     |        48 MiB |
 | npm tarball                       |  90,000 bytes |
 | npm unpacked package              | 300,000 bytes |
@@ -50,6 +50,7 @@ npm run format:check
 npm run lint
 npm run typecheck
 npm test
+npm run perf:check
 npm run perf:audit
 npm run bench -- --iterations 10 --json
 npm run bench -- --startup-only --json

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-browser-use",
-  "version": "0.9.5",
+  "version": "0.9.6",
   "description": "Opinionated browser-use for the Pi coding agent, powered by chrome-devtools-mcp (not Playwright). Fresh headless by default, authenticated persistent profile opt-in, CLI-first policy bundled.",
   "keywords": [
     "automation",
@@ -51,8 +51,10 @@
     "typecheck": "tsc -p tsconfig.json --noEmit",
     "prepublishOnly": "npm run build",
     "bench": "node scripts/bench.mjs",
-    "test": "node --test test/*.test.mjs && npm run perf:check",
+    "test": "node --test test/*.test.mjs",
+    "preperf:audit": "npm run build",
     "perf:audit": "node scripts/performance.mjs",
+    "preperf:check": "npm run build",
     "perf:check": "node scripts/performance.mjs --check",
     "pretest": "npm run build"
   },