v3.8.3

[3.8.3] - 2026-07-07

Patch [behaviour] — Tier D bundle D9 + D10: new browser-tools skill (Edition + Supervisor byte-identical mirror) + welcome-baseline $comment note documenting Copilot Memory user preferences on Business/Enterprise plans.

User signal confirmed both preconditions the Tier D menu was holding these items on: (a) enterprise plan in use, unblocking D9's Business/Enterprise Copilot Memory user-preferences awareness; (b) concrete use case named for browser tools — bot-avoidance across multiple projects + design validation — unblocking D10's new skill.

Source: Tier D items D9 + D10 from Alex_ACT_Supervisor/docs/proposals/tier-d-implementation-menu-2026-07-07.md.

Added

• .github/skills/browser-tools/SKILL.md (~200 lines) — new skill guiding when to reach for VS Code 1.127+ browser tools instead of fetch_webpage. Covers three operational patterns: (1) bot-protected content read (CloudFlare, PerimeterX, Akamai, Datadome), (2) design validation via screenshot_page, (3) interactive site behind consent gate. Toolset reference for all agent-invocable browser operations (open_browser_page, screenshot_page, click_element, read_page, type_in_page, hover_element, handle_dialog, drag_element, run_playwright_code, navigate_page). Safety section: external URL trust boundary + screenshot content leakage + credential handling (NEVER type secrets via type_in_page) + enterprise policy interaction (BrowserChatTools, ChatAgentNetworkFilter) + consent-gate discipline (don't accept ToS on user's behalf). Cost section: browser tools heavier than fetch_webpage, upgrade only when needed. When NOT to Fire: static HTML, trusted docs sources, API endpoints. Byte-identical Supervisor mirror.

Changed

• .github/config/welcome-baseline.json ($comment only, settings block untouched) — Category (3) Copilot chat features gained 5-line sub-note documenting Copilot Memory user preferences (GH GA 2026-06-02 for Business/Enterprise plans) as a distinct layer from this brain's memory-triggers.instructions.md. Platform Copilot Memory persists short-term chat context on GitHub-managed infra; brain memory-triggers govern deliberate persistence to /memories/, HANDOFF.md, and ../Alex_ACT_Memory/. Both fire concurrently without conflict.

Brain contract

Unchanged: min_extension_version stays 9.5.1, brain_subtrees stays [.github], marker_schema stays v2, manifest spec stays 1.4. Only edition_version (3.8.2 → 3.8.3), generated_at, and skills count (37 → 38) change in the manifest.

Heir upgrade

Standard /upgrade (Extension ACT: Upgrade Brain). No --allow-major needed. Heirs picking up v3.8.3 get the new browser-tools skill available for description-match invocation + the welcome-baseline $comment refresh. No rule changes; behavior gate for existing rules unchanged.

Falsifier

2026-10-07 (90 days) per skill's own ## Would Revise If:

• Browser tools fire on tasks where fetch_webpage would have worked ≥3 times in a quarter → tighten "When to Fire" criteria.
• Bot-protected content still fails on browser tools ≥1 time → surface the failure mode in Pattern 1.
• Enterprise policy BrowserChatTools blocks tools entirely for a heir's org → add "not-applicable" branch to the skill.
• Safety incident (screenshot leak, credential typed to wrong field, ToS accepted without authorization) → expand Safety section.
• Design-validation output discipline slips (agent claims "looks good" without evidence) → tighten Pattern 2 evidence-required rule.