$ cat ai-agents/week/2026-10-12.log
the week in review · Oct 12 – Oct 18, 2026
Claude Code RCE via malicious git executable
By RepoJournal · composed from the cited sources · human-reviewed weekly · methodology
Cline shipped v3.0.71 to v3.0.73 for people running plugins on Windows without a JS runtime, and restored GPT-6 on OpenAI endpoints.
Uncontrolled Search Path Enables Arbitrary Code Execution via Malicious Git Executable anthropics/claude-code
The advisory describes an uncontrolled search path: a hostile git executable on the path gets executed by Claude Code. The fix is a version bump, and it matters because the attack surface is the directory and checkout you start the tool in.
CLI v3.0.71 cline/cline
by github-actions[bot]
Cline moves plugin loading into the hub process, so plugins run without a JavaScript runtime on your PATH. A plugin that fails now prints a warning naming it and the error instead of silently dropping its tools, and cline config tools reports what loaded.
Desktop v0.0.48 cline/cline
by github-actions[bot]
GPT-6 models such as gpt-6-sol and gpt-6.1-sol work on OpenAI and Azure OpenAI-compatible endpoints. Every request was rejected before because Cline sent max_tokens rather than max_completion_tokens. With no model chosen, the Cline provider defaults to Claude Sonnet 5.5 again.
feat(core): run plugins only in workers; drop the in-process mode cline/cline
The in_process execution mode is gone, along with the fallback a host used when it shipped no worker bootstrap. That fallback turned a loud plugins-do-not-load into a silent failure, so a misconfigured host now fails visibly rather than running plugin code on the main thread.
fix(providers): reject malformed GOOSE_MODE across ACP providers aaif-goose/goose
A malformed GOOSE_MODE now fails loudly across claude-acp, copilot-acp, amp-acp, pi-acp and claude-code. Previously a bad value silently selected permissive Auto mode, which meant a typo in your config could leave approval gating off; Auto now applies only when the setting is absent.
agent-loop: not infer or steer while a tool approval is pending aaif-goose/goose
The agent loop no longer re-calls the provider while a tool approval is pending, and a queued steer no longer lands on that round. Operators gating destructive tools get the approval gate they configured instead of a request carrying an unanswered tool_call.
Support Windows clipboard text paste shortcuts in the TUI openai/codex
On Windows, Ctrl+V and Shift+Insert now take the clipboard paste path instead of attempting image paste only, so text-only clipboards stop erroring in the TUI. Composer selections also honour copy-on-select on release, with X11 PRIMARY publication when enabled.
fix(desktop): bundle arm64 binaries for windows arm anomalyco/opencode
Alongside: the Windows ARM64 desktop build no longer bundles x64 native binaries, opencode stops listing Azure resources at startup, Gemini CLI keeps loading extensions past one failed allowlist check, and Codex isolates its arg0 helper directories per platform.
$ ls ai-agents/week/ # the briefings behind this review
Keep up with Agentic Coding in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.
Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.