prism-mcp-server 20.3.0 → 20.3.1

package/README.md

@@ -61,6 +61,41 @@ features.
 <details>
 <summary>Release history (optional)</summary>
 
+## What's New in v20.3.1
+
+### Prism Browser Reports Real Failures
+
+`prism browser` could not fail a test. `eval 1 === 2` returned `status: ok`
+with exit code 0, a page serving HTTP 500 reported `status: ok`, and console
+errors and uncaught page exceptions were discarded entirely. This release adds
+assertions — `assert-text`, `assert-visible`, `assert-count`, `assert-url`,
+`assert-title`, `assert-eval`, `assert-no-page-errors` — that return
+`status: failed` and a non-zero exit. `open` now reports `http_status` and
+fails on 400 or higher, screenshots are validated rather than assumed, and
+`eval` returns native JSON with its type instead of a Python `repr`.
+
+The fingerprint layer had never been applied: a wrong keyword argument made
+the stealth library throw on every launch — 1,139 failures and 0 successes
+since April — while the runner reported it as active. It is fixed, and a layer
+that cannot be applied now fails loudly. The headless build no longer
+advertises itself through `navigator.userAgentData` or the `Sec-CH-UA` header,
+and a patch that corrupted `Object.getOwnPropertyDescriptor` on every page
+under test has been removed. These remain best-effort test aids, not a
+guarantee against bot detection.
+
+`--local-only` now actually isolates: WebSocket, EventSource, WebRTC and
+`sendBeacon` egress bypass request routing and were never blocked, and service
+workers were allowed through. `--cleanup` was a no-op in the two modes agents
+use. Site isolation, phishing detection and popup blocking are no longer
+disabled by default, since these profiles hold live authenticated cookies.
+
+New for test runs: `--ephemeral-profile` and `--storage-state` for hermetic
+authenticated flows, `pages`/`switch-page` so OAuth popups are reachable,
+`--fail-fast`, `--fast`, `--trace`/`--video`/`--har`, and
+`profiles --prune-older-than` for profile maintenance.
+
+---
+
 ## What's New in v20.2.7
 
 ### Session Saves Survive Agent Restarts
@@ -81,12 +116,12 @@ to the patched 8.5.23 release.
 
 ### Hybrid Memory Search (Portal Tier)
 
-`session_search_memory` on Synalux-backed installs now fuses semantic
-similarity with exact-term lexical matching (weighted reciprocal-rank
-fusion). On blind probes against a real 8.5k-entry corpus this lifted
-top-1 retrieval from 45% (semantic alone) to 59%; exact identifiers such
-as TPNs, function names and error strings now rescue queries that
-embeddings blur. Results say how they were found — `hybrid retrieval`
+`session_search_memory` on the portal tier (Synalux-backed installs) now
+fuses semantic similarity with exact-term lexical matching via weighted
+reciprocal-rank fusion. Measured on blind probes against a real
+8.5k-entry corpus: fused retrieval was **never worse** than semantic
+alone at top-5, and exact identifiers — TPNs, function names, error
+strings — now rescue queries that embedding similarity blurs. Results say how they were found — `hybrid retrieval`
 headers, per-hit `sem#/lex#` arms — and a lexical-only rescue is labelled
 `exact-term match` instead of pretending to a similarity score. Local
 SQLite installs keep pure vector search; hybrid needs the portal's

package/docs/prism-browser.md

@@ -18,16 +18,100 @@ so a separate app or DMG is not required.
   scripts, allowing deterministic feature flags, fixtures, capability shims,
   or instrumentation for localhost tests.
 - **A constrained injection boundary.** Injection requires `--local-only`,
-  which rejects public navigation and non-loopback subrequests.
+  which rejects public navigation, non-loopback subrequests, service workers,
+  and WebSocket/EventSource/WebRTC egress.
 - **Private audit records.** The runner stores a local audit trail with private
   filesystem permissions and removes URL credentials, query strings,
-  fragments, common email/phone patterns, and injected source text.
+  fragments, common email/phone patterns, record identifiers in URL paths, and
+  injected source text.
+- **Failure signals instead of silence.** Console errors, uncaught page
+  exceptions, and failed requests are captured and attached to command output.
+  An HTTP status of 400 or higher, an empty screenshot frame, and a failed
+  assertion each produce a non-zero exit.
 
 These are orchestration and safety benefits. Prism Browser does not replace
 Playwright Test: use raw Playwright when you need its complete fixture,
 assertion, trace, project, or parallel-worker APIs. Compatibility patches are
 best effort and are not a CAPTCHA-bypass guarantee.
 
+## Assertions and exit codes
+
+A command reports `status: "ok"` only when it actually succeeded. Anything else
+(`failed`, `error`, `timeout`) marks the run, and `pipe` exits non-zero.
+
+```bash
+printf '%s\n' \
+  'open http://127.0.0.1:3000' \
+  'assert-title Dashboard' \
+  'assert-text #status Ready' \
+  'assert-count [data-row] 12' \
+  'assert-no-page-errors' \
+  | prism browser --headless --fast --fail-fast pipe
+```
+
+`eval` returns the native JSON value with its type, so results are machine
+readable:
+
+```json
+{"status": "ok", "result": {"a": 1}, "type": "dict", "serializable": true}
+```
+
+`assert-eval` is the assertion form — `eval` alone never fails a run, because a
+falsy expression is a legitimate result. Use `--fail-fast` for test runs so a
+failed navigation cannot be followed by commands that silently target the
+previous page.
+
+A trailing backslash continues a command onto the next line, which makes
+multi-line JavaScript and multi-line input text expressible. Quoted arguments
+are parsed with shell-style quoting, so `type "div > .cell" "two words"` works.
+
+## Hermetic runs
+
+Persistent profiles are convenient for interactive work and wrong for tests:
+state carries between runs and makes them order dependent. For test runs use
+
+- `--ephemeral-profile` — a throwaway profile directory, removed on exit.
+- `--storage-state PATH` — seed cookies and localStorage from a Playwright
+  storage-state file, so authenticated flows skip the login UI.
+- `save-storage PATH` — write the current state back out.
+- `--fast` — skip the human-latency emulation (per-character typing and
+  inter-action pauses), which otherwise dominates a test's wall clock.
+
+`--trace PATH`, `--video DIR`, and `--har PATH` record Playwright artifacts for
+a failing run.
+
+## Multiple pages
+
+A popup (OAuth, payment, print preview) opens a new page. `pages` lists them,
+`switch-page N` makes one active, and `close-page` disposes of it. Without
+switching, commands continue to target the original page.
+
+## Fingerprint patches
+
+`--stealth full` applies playwright-stealth, a CDP
+`Emulation.setUserAgentOverride` carrying full `userAgentMetadata`, and a
+supplementary init script. The CDP override is what keeps `navigator.userAgent`,
+`navigator.userAgentData`, and the `Sec-CH-UA` request headers consistent;
+rewriting headers through request interception does not work, because Chromium
+re-adds client hints after interception.
+
+Applied layers are verified rather than assumed. The runner probes the page for
+UA/brand/platform/webdriver agreement at startup and again after the first real
+navigation, and `--stealth full` fails with an actionable error when a requested
+layer cannot be applied. Pass `--allow-degraded-stealth` to proceed anyway, or
+`--stealth light` to skip the library layer deliberately. The `fingerprint`
+command reports the current state.
+
+These patches do not disable site isolation, client-side phishing detection, or
+popup blocking. Profiles hold live authenticated cookies, and trading away those
+mitigations for a fingerprint delta is the wrong exchange.
+
+## Profile maintenance
+
+Persistent profiles accumulate. `prism browser profiles` lists them by size and
+age; `--prune-older-than DAYS` reports what is stale and deletes it only when
+`--yes` is also passed.
+
 ## Install the local runtime
 
 ```bash

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "prism-mcp-server",
-  "version": "20.3.0",
+  "version": "20.3.1",
   "mcpName": "io.github.dcostenco/prism-coder",
   "description": "Prism Coder — Cognitive memory + tool-calling intelligence for AI agents. Mind Palace persistent memory (BFCL Gold Certified, 100% Tool-Call Accuracy, 114 Agent Skills, PHI Guard, Tier Enforcement, Prompt-Based Skill Routing, Zero-Search HDC/HRR retrieval, HRR Semantic Drift Detection across BCBA/Coding/AAC domains, HIPAA-hardened local or subscription-gated Synalux storage, SLERP-optimized GRPO alignment) plus the prism-coder 1.7B–32B open-weights LLM fleet.",
   "module": "index.ts",