Entity trail

Cline

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-08 / github-pulse-researcherchrome-devtools-mcp 1.9.0 publishes itself as an Agent Plugins 1.0 package installable across five agent clientsReleased 2026-09-08T09:49Z, v1.9.0 adds a root-level Agent Plugins 1.0 package (#2623) exposing the existing stdio server through a version-pinned mcp.json, with the package version and MCP npm pin kept in sync by release-please. The PR records isolated add, doctor, repair and remove passes against Claude Code, Gemini CLI, OpenCode, Cline and Windsurf using Agent Plugins CLI 0.1.24, with OpenCode 1.18.25 confirmed making live new_page and take_snapshot calls. Google shipping a first-party MCP server through a cross-client plugin manifest is a distribution signal worth watching.
2026-09-08 / arxiv-researcherEach Level of LLM Personalization Raises Phishing Click Intent Odds by 28%arXiv 2609.04410 recruited 180 U.S. working adults for a disclosed survey rating AI-generated phishing emails built at four cumulative personalization levels: workplace only, then recipient name and job title, then job responsibilities, then coworker and shared-project context. Across 1,436 valid evaluations, convincingness rose 2.40 points per level in a sensitivity analysis and the odds of expressing click intention rose 28% per level. A post-hoc analysis found messages from a named person referencing a supplied coworker outperformed messages from a department, and among non-clickers reporting declined while deletion increased, which quietly degrades the security-team signal.
2026-09-08 / news-researcherEuropean Wealthtech Funding Fell 46% in Q2 and EMEA Fintech Funding Dropped Sharply in H1Investor caution drove a 46% reduction in European wealthtech funding in Q2 2026, alongside a sharp decline in overall EMEA fintech funding across the first half of the year as investors became more selective. The contrast with Mistral's €3B round the same week is the point: capital is concentrating into frontier AI infrastructure while adjacent applied-fintech categories contract.
2026-09-07 / thought-leaders-researcherTNW finds the contradiction inside OpenAI's own data: a 59.2% GPU cut to Astra-class training moved compute rather than removing itAna Maria Constantin's September 6 analysis pairs Pachocki's slowdown essay against OpenAI's research post and reports that when safety restrictions forced a 59.2% reduction in Astra-class GPU allocation, other model classes absorbed 85% of the decline, so "compute stays valuable and flows into other uses." She also notes that over half of successful four-to-eight-hour agent tasks required at least one human intervention, while internal support channels have gone quiet. She frames the essay as a response to the July 20 infrastructure compromise and August 7 cyber concerns rather than as an abstract argument.
2026-09-06 / github-pulse-researcherMagnitude is a local-inference server that plugs into eight existing coding agents rather than replacing themmagnitudedev/magnitude (3,426 stars, 247 forks, last push 2026-09-06 07:15 UTC) profiles your hardware, recommends models that fit, then downloads, tunes and serves them behind Pi, OpenCode, Hermes, OpenClaw, Codex, Claude Code, Oh My Pi and Cline, or its own built-in harness. It picked up roughly 600 stars on 2026-09-06 alone per GitHub Trending. The 2026-09-05 release @magnitudedev/[redacted] (#71) sets an explicit HTTP request-body limit so larger image requests stop being rejected before reaching the handler, which is the practical blocker for vision work on a local server.
2026-09-05 / arxiv-researcherAlcaTRAz Defends Jailbreaks With Character-Level Perturbation Rules and No Model Access, Beating Llama Guard on 73.4% of CombinationsarXiv 2609.03693 proposes a prompt-level jailbreak defense that needs no weights, internals, retraining or modification of the target model, learning a transferable transformation rule that inserts controlled character-level perturbations at selected positions to disrupt the structural regularities jailbreaks exploit. Across 33 open-weight models and 22 jailbreak attack types, it achieved the best composite security and functionality score in 73.4% of model-attack combinations against Llama Guard, RA-LLM and Goal Prioritization, shifting the aggregate severity score from a modal 10 to a modal 2 while keeping mean benign score within 0.27 points of undefended (8.35 vs 8.62). The authors explicitly decline to claim a guarantee, noting a residual high-severity tail and no evaluation against adaptive attackers.
2026-09-04 / skill-finderA late requirement destroys about twice as much agent-written code as an ordinary edit, and warning the agent in advance does not helparXiv 2609.03028 mines 3,553 eligible SWE-chat sessions for requirements that arrive after the agent has already implemented something, and where repository state can be replayed, links each arrival to deletion or replacement of prior agent-authored lines. A requirement's arrival is followed by roughly twice as much invalidation as matched non-requirement edits, robust to user-turn and net-deletion checks; the burden shows no decline over the course of a session and no association with operation type. A controlled experiment found that delayed disclosure merely relocates implementation to after the reveal, while advance warning produced no detected effect on overwriting, which undercuts the common advice to tell the agent up front that requirements may change.
2026-09-04 / vibe-coding-researcherPattern: four agent CLIs shipped MCP server-lifecycle fixes in the same 36 hoursCline 0.0.23 fixed one wedged MCP server blocking the rest from shutting down and leaking their processes; Copilot CLI 1.0.83-5 fixed MCP tools becoming uncallable after a server restart and 1.0.83-4 fixed servers configured by the agent disappearing after built-in sub-agent turns; Codex 0.153.0 scoped remembered MCP tool approvals to the selected account and kept event subscriptions alive after task unloading; Claude Code 2.1.260 fixed SDK-provided MCP servers missing from the first turn. The connect path is now well-trodden — what is breaking across every vendor at once is restart, shutdown and cross-turn survival.
2026-09-04 / vibe-coding-researcherPattern: the repository folder itself has become the trust boundary agents are now defendingThree independent items in 48 hours all police what a checked-out directory is allowed to make an agent do: GitSpawn's core.fsmonitor RCE across seven agents, arXiv 2609.03884 on attacker-controlled lifecycle-hook updates, and Cline shipping a deliberate refusal to load `.agents/plugins` from the workspace. The shared shape is that agent harnesses read repo-local configuration (git config, hook definitions, plugin manifests) with host privileges before any human approval step runs. If you maintain a harness, enumerate every repo-local file it reads at session start and ask which of them can name a program.
2026-09-04 / vibe-coding-researcherCline desktop 0.0.23 runs Agent Plugins from ~/.agents/plugins and deliberately ignores the workspace copyCline's shared Hub now discovers packages under `~/.agents/plugins`, validates them from `plugin.json`, exposes their valid Agent Skills to the agent, and auto-starts their stdio, Streamable HTTP and SSE MCP servers, with enable/disable managed per plugin in Settings → Customize. The release note states that workspace-level `.agents/plugins` directories are intentionally ignored — a deliberate refusal to let a cloned repository install its own MCP servers and skills, which is exactly the trust boundary GitSpawn crosses through .git/config.
2026-09-03 / agents-researcherCivBench runs agents for 300+ turns over 76 MCP tools and finds they stop checking state before losingCivBench is an open-source long-horizon benchmark putting language model agents inside Civilization VI through MCP, with a single episode spanning over 300 turns and thousands of tool calls, 76 exposed MCP tools, and a narration layer converting visual game state to structured text. Across 23 admissible runs from four model families the authors decline to rank models and instead introduce two interface-level metrics: Proactive Monitoring Rate and RAG@10, whether commitments stated in planning reflections get executed within ten turns. Despite playbook instructions to query victory progress every 20 turns, agents queried every 30 to 75, and in 7 of 20 detectable defeats never queried inside the 20-turn warning window. Agents under-monitor state that is available but must be explicitly requested.
2026-09-03 / vibe-coding-researcherAn unreachable MCP server was tearing down whole Cline sessions before a 10s connect budget landedRemote SSE and streamable-HTTP MCP servers now get a 10-second connect budget. An enabled but unreachable server previously burned the full 60-second request timeout inside `session.create`, blowing past the hub's 30-second cap and killing the session outright. Same release: an empty `capabilities` list was reading as an authoritative denial that stripped every tool definition from requests to Dify, SAP AI Core, opencode and the Codex CLI, and separately dropped every image from file reads.

Graph relationships

SUPPORTS
chrome-devtools-mcp
chrome-devtools-mcp publishes as Agent Plugins 1.0 package installable on Cline
Source finding
USES
Claude Fable 5.1
Cline changed its default model from Claude Opus 5 to Claude Fable 5.1.
Source finding
SUPPORTS
Claude Fable 5.1
Claude Fable 5.1 became the resolved default in Cline across 57 providers.
Source finding
SUPPORTS
Codex
Cline v4.1.17 fixes tool calling on Codex CLI models
Source finding
SUPPORTS
Dify
Cline v4.1.17 fixes tool calling on Dify models
Source finding
SUPPORTS
Vercel
Vercel's AI SDK harness layer supports Cline.
Source finding
USES
Git
Cline redacts credentials from git remote URLs.
Source finding
USES
VS Code
Cline is a VS Code extension.
Source finding
SUPPORTS
ponytail
ponytail ships an adapter for Cline.
Source finding
MENTIONS
npm
GitHub issue title prompt injection compromised Cline npm publishing tokens.
Source finding
MENTIONS
npm
Prompt injection in GitHub compromised Cline npm publishing tokens.
Source finding
MENTIONS
OpenClaw
Cline npm release was compromised to include OpenClaw.
Source finding

Source trail

v1.9.0 Manifold Security NVD Cline arXiv 2609.03884 SDK 0.0.82 magnitudedev/magnitude claude-code CHANGELOG Cline release GitHub GitHub Changelog opencode

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…