SI Tools 2026 10 30 Your guard never ran

SI Tools 2026 10 30 Your guard never ran On a Windows machine without Git Bash, GSD's installer passed while its hooks pointed at WSL's bash. GSD's own PR calls them "hooks that never run." This week's theme is guards that were set up but did nothing. Hooks that didn't fire, across three projects: Open GSD Core 1.19.0 now fails loudly in that Windows case. Claude Code 2.1.314 fixes Bash hooks skipping Monitor-tool commands, and 2.1.310 fixes Windows hooks waving tool calls through when Git Bash was missing. The hookify plugin's block rules "previously failed silently." A marketplace cleanup quietly turned off a security reviewer: security-guidance stopped reviewing for some users and stays off until /plugin enable. blockedMarketplaces doesn't stop git-subdir plugins. code-review 1.0.1 no longer comments on PRs unless you pass --comment. Claude Code 2.1.314 closes policy holes: sessions could start without policy when the server was unreachable, claude -p exited 0 when it never started, and there's a new blockWritesOutsideWorkingDirectories setting. In 2.1.312 the safety_event telemetry event is gone again. GSD's secret guard let .env reads through wrappers (timeout, nice, env -i, exec, grep -if.env). Also: an undocumented Claude Code block on subagents writing SUMMARY.md and similar files. Grok Build and Grok Bot move their approval lines: in Grok Build, plan mode finally means plan mode and worktrees inherit approvals. In Grok Bot, sends you asked for go out after Auto-review, and cards stay open 30 minutes. Version pins (previous in parentheses): Claude Code 2.1.314 (2.1.309); npm stable 2.1.308 unchanged; 2.1.313 was skipped Open GSD Core 1.19.0 (1.18.0) Buildomator 5.0.4 (5.0.3) Superpowers 7.0.1 (no change) security-guidance 2.0.17 (2.0.16) code-review 1.0.1 (new pin) Grok Build 1.0.70 (1.0.62) Grok Bot 0.78.1 (0.75.2) X posts referenced: Confirmed: @ClaudeDevs 32 MB to 128 MB requests https://x.com/ClaudeDevs/status/2115923971... Confirmed: @ClaudeDevs MCP Tasks and Elicitations https://x.com/ClaudeDevs/status/2113698001... Confirmed (as a statement): @bcherny token burn, no single root cause https://x.com/bcherny/status/2115592371224... Unverified: @bcherny CLAUDE_CODE_SUBAGENT_EFFORT "landed" https://x.com/bcherny/status/2115859487068... Confirmed: @trq212 left Anthropic https://x.com/trq212/status/21138440519380... Partly confirmed: @theagenticdaily 2.1.310 "fails closed" https://x.com/theagenticdaily/status/21143... Confirmed: @JustJorshin deny rule "isn't a security boundary" https://x.com/JustJorshin/status/211545100... Partly confirmed: @0xgloob plugin code bypasses the sandbox https://x.com/0xgloob/status/2115234851884... Confirmed: @DegenOfWallSt stable and Homebrew lack 2.1.309 https://x.com/DegenOfWallSt/status/2114494... Confirmed: @bot Vault https://x.com/bot/status/2113717946019397910 and @elonmusk https://x.com/elonmusk/status/211379618368... Claim only: @elonmusk Bot places phone calls https://x.com/elonmusk/status/211436536106... Claim only: @elonmusk other messaging platforms "coming soon" https://x.com/elonmusk/status/211475307016... Claim only: @elonmusk 100+ features and fixes a week https://x.com/elonmusk/status/211475173902... Partly confirmed: @elonmusk no free tier https://x.com/elonmusk/status/211592160065... Partly confirmed: @mattyp What's New chapters https://x.com/mattyp/status/21151118973976... X posts are treated as leads only. Each was checked against primary sources (changelogs, docs, release notes, npm) and labeled. AI-generated audio deep dive (NotebookLM).