Add prepared build artifacts, browser host and the build and
purge commands for the CLJS renderer benchmarks.
Each preparation attempt gets a build ID and a retained directory with
its manifest, log, Node case catalog, browser bundle, renderer files
and host page. Ready builds load their matching catalog in process and
serve only their own directory through an owned loopback host.
Own Chromium through a bounded process group, verify the compiled build
identity before execution and retain failed preparations for diagnosis.
Wire the diagnostic pilot to the same build, catalog, server and
browser APIs.
Add real boundary tests for artifacts, processes, servers, browsers and
CLI parsing, and register them with the performance test runner. Rename
the benchmark lint and test scripts to match the suite.
AI-assisted-by: deepseek-flash
./build and pnpm run build:wasm produced no artifacts, dropped extra
Cargo arguments, and watch compiled outside the lock without copying.
Build now runs cargo build and copy_target_artifacts under the
existing lock. Cargo arguments survive as an array, and watch re-entry
no longer duplicates profile flags.
Destinations accept environment overrides so callers can stage
artifacts outside the public paths.
AI-assisted-by: deepseek-flash
An operator writing `PENPOT_TENANT=ACME` and one writing `acme` name the
same tenant: the value is compared as an exact string in the keys and
rows it lands in, and one uppercase letter would otherwise split one
tenant into two.
The value is normalized when the configuration is read, so nothing
downstream has to care about case, and the validated form accepts only
lowercase labels from then on.
AI-assisted-by: deepseek-v4.1-flash
* 🐛 Fix ChatGPT tool discovery with the submit_feedback tool
ChatGPT compiles the regex patterns advertised in tool input schemas
with Python, which rejects Unicode property escapes such as \P{Cc}. The
submit_feedback subject pattern used one, so ChatGPT dropped the whole
tool list ("action discovery failed") on deployments that register the
tool.
The subject pattern now spells out the Cc range as a character class,
which accepts the same input. A test asserts that no advertised pattern
uses \p{...} or \P{...}.
AI-assisted-by: claude-opus-5-5
* 🐛 Fix MCP plugin test stub for the version check
plugin.ts checks version compatibility with extractMinorVersion, but the
tests loaded it with a require stub that returned the same dummy object
for every import, so the check threw once the UI initialized.
The plugin.ts tests now share a stub that supplies the real
VersionCompatibility, ErrorUtils and TaskHandler modules, stubs only the
code execution handler and rejects unexpected imports. TaskHandler is
transpiled for the sandbox because Node's type stripping cannot load its
parameter properties.
AI-assisted-by: claude-opus-5-5
* 🐛 Fix Management tests for backend-confirmed setters
Management setters save to the backend first and update the object only
when the save succeeds, so seven live tests failed:
- setter tests now wait for the new value before reading it
- createProject checks fileCount by listing the project again, as
fileCount is the count from when the project was listed
- project moveTo no longer calls remove on the target team, which teams
do not have; the live CI driver deletes temporary teams
- renaming the open file waits for its restore, which made the next
navigation test fail
The README notes on setters, fileCount and team cleanup match this.
AI-assisted-by: claude-opus-5-5
