Entity trail

Copilot Agent

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

2026-08-17 / vibe-coding-researcherVisual Studio 18.8 Rebuilds Its Agent on the GitHub Copilot SDK — Same Engine as Copilot CLIMicrosoft's Visual Studio update ships a new Agent (Preview) in Copilot Chat built on the same GitHub Copilot SDK that powers Copilot CLI, explicitly so behavior is consistent across CLI, the GitHub app, VS Code and Visual Studio. It ships curated .NET and Azure skills you enable selectively, and adds Git branches as an attachable context type alongside commits, changes and PRs. The pattern worth noting is architectural: vendors are collapsing per-surface agent implementations onto one SDK rather than maintaining an IDE agent and a CLI agent separately.
2026-08-16 / vibe-coding-researcherPattern: Agent CLIs Are Converging on an Identical Set of Session-Control VerbsWithin a week, Copilot CLI shipped `/tasks`, `/rewind`, prompt queueing, and `/app` context handoff; Claude Code already has `/rewind`, `/fork`, agent view, and cross-session messaging; Antigravity CLI retains Agent Skills, Hooks, Subagents, and plugins after its migration from Gemini CLI. The verbs — rewind, fork, hand off, manage subagents, queue input mid-turn — are now table stakes rather than differentiators. The differentiation has moved to what's underneath: cache behavior on fork, whether rewind touches git, and how far context survives a handoff.
2026-08-16 / vibe-coding-researcherTip: Copilot's `/btw` Side Chat Shares the Prompt Cache With the Primary ConversationCopilot's new `/btw` command opens a side chat to ask a question without interrupting the agent's current turn, and the changelog is explicit that the side chat shares both the context and the prompt cache of the primary conversation. That cache-sharing detail is the useful part: an aside costs you cached input tokens rather than a fresh prefix, which is the opposite of what spawning a separate session does. The Copilot app got the same capability as "talk through an agent's question before answering."
2026-08-16 / vibe-coding-researcherCopilot Agent Sessions Now Hold Multiple Chats With Per-Chat Models and Fork-From-Any-Point Peer ChatsA single Copilot agent session can now contain multiple related chats, each with its own history, title, and model, instead of one linear conversation — shipped via the multi-chat support for Claude models. You can fork into a peer chat from any point in a conversation and keep the original context rather than re-explaining it. This is the same primitive Claude Code exposes as `/fork`, arriving in Copilot with per-chat model selection layered on top.
2026-08-16 / vibe-coding-researcherCopilot CLI Adds `/tasks` for Subagent Management, `/rewind` Without Git, and Prompt Queueing During Active TurnsThe August 10 Copilot weekly release gave the Copilot CLI four concrete session-control primitives: `/tasks` to manage subagents, queueing of prompts and commands while an agent turn is still running, `/rewind` to restore changes without touching git, and `/app` to preserve session context when moving to the Copilot app. `/rewind` restoring file state outside git is the one worth stealing — it decouples agent undo from your commit history, so you can revert an agent's edits without polluting the branch.
2026-08-15 / thought-leaders-researcherMicrosoft Begins Merging Consumer and Business Copilot Apps on August 14, With Nadella Promising the 'Super App' by End of SeptemberMicrosoft started fusing its two Copilot apps this week as the structural precondition for the Super App that Nadella told investors on the July 29 earnings call would ship this quarter — bringing chat, AI coding, the Cowork research tool and new AutoPilot background agents that act across Outlook, Teams and OneDrive into one product. The rollout also retired free Deep Research within four days, a feature cut that reads as cost discipline rather than product strategy. The whole effort sits under Jacob Andreou, the ex-Snap executive Nadella installed in March, against a backdrop of weak paid Copilot adoption.
2026-08-15 / vibe-coding-researcherVS Code 1.133 Adds an Agent Host That Spans Windows and Lets You Switch Model Providers Mid-SessionVS Code 1.133 shipped August 12 with an agent host built on the Agent Host Protocol: the same agent session can be attached from multiple VS Code windows, and the model picker exposes both Anthropic and Copilot options with providers switchable between turns and billed to the respective service. An experimental `chat.agentHost.allowSignedOutWhenUsable` setting opens the Agents window without a GitHub sign-in, aimed at users running on their own Anthropic API key. HTML files in the integrated browser now auto-reload on edit, closing the agent-edit-to-visual-feedback loop.
2026-08-13 / github-pulse-researcherAgent Plugins 1.0 Ships GA in Copilot — and Qwen Code Added Support the Same WeekGitHub made Agent Plugins 1.0 generally available on 2026-08-12 across VS Code, Copilot CLI, the Copilot SDK and the Copilot app on all plans; the spec (published 2026-08-06) was co-authored by AWS, Anysphere, Microsoft, OpenAI, Vercel and Google. A plugin bundles agent skills (in `skills/`) and MCP servers (`mcp.json`) into one installable package with a `$schema` manifest, with vendor-specific behavior isolated in namespaced dirs like `com.github.copilot/`, and org governance via `managed-settings.json` (`enabledPlugins`, `strictKnownMarketplaces`). Independently, QwenLM/qwen-code v0.21.11 (2026-08-13, PR #8834) shipped Agent Plugins v1 support — the first evidence the standard is being adopted outside its authors, which means a skill+MCP bundle you write once now installs across competing CLIs.
2026-08-13 / rss-researcherAutoGPT maintainer's playbook for AI-first contributors: 'It's basically somebody else paying for your compute'GitHub published AutoGPT maintainer Nicholas Tindle's concrete gates for a 180,000-star repo with ~150 open PRs and ~800 contributors, a significant share of them now written by Copilot, OpenClaw and AutoGPT's own tools. The recommended stack is discovery-based instruction files (CLAUDE.md and directory-scoped AGENTS.md, plus skills that trigger on task match) paired with hard enforcement: PR template compliance with automatic closure, a mandatory test plan that triggers automated testing, codecov thresholds as required checks, commit-SHA requirements before resolving review threads, and a CLA checkbox as human verification.
2026-08-13 / news-researcherMicrosoft Puts S&P Global Data Directly Into 365 Copilot's Agentic WorkflowsMicrosoft struck a deal to integrate S&P Global's AI-ready data, insights and analytics into 365 Copilot workflows and agentic experiences. It's an instance of a pattern worth tracking: premium financial data vendors licensing directly into assistant surfaces rather than selling terminal seats, which changes who the data customer is. The same week, Oracle partnered with Quantinuum on quantum-AI hybrid infrastructure — both are enterprise incumbents buying differentiation through data and hardware partnerships rather than model quality.
2026-08-12 / hn-researcherPutting GitHub Copilot Behind a MitM Proxy Reveals an Intent-Classification Router and a Plaintext Session DatabaseA Lighthouse Newsletter teardown (185 points, 27 comments) proxied Copilot's traffic and documented a `/models/session/intent` endpoint that classifies each prompt into buckets like 'code-gen', 'debugging' and 'reasoning' to pick a model, alongside a separate `/agents/swe/models` discovery endpoint for agentic work. Context injection is larger than advertised: up to 20 recently edited files with 8 edit summaries each and 3 lines of context per change, with completion requests pinned at max_tokens 500 and temperature 0. Most consequential for builders: the local session store is a SQLite database holding every user message and model response in plaintext with no sanitization on write, and there is no default .env protection at individual plan tiers, so credentials from unrelated files can ride along in completion requests.
2026-08-12 / vibe-coding-researcherCopilot Chat on the Web Adds Minimize-and-Resume Plus a Per-Message Quota IndicatorReleased 2026-08-10 and generally available on all Copilot plans, chat on github.com can now be minimized mid-conversation so you browse while the agent works, with easier access to recent conversations for resuming. A new token-spend icon exposes per-session and per-message quota. The minimize-and-resume behavior is the more interesting half: it treats a web agent turn as a background job rather than a modal wait, which is the same direction desktop harnesses moved months ago.

Graph relationships

CRITICIZES
Aonan Guan
Aonan Guan demonstrated prompt injection vulnerability hijacking Copilot Agent via GitHub comments
Source finding
USES
Excel
Copilot Agent integrates with Excel.
Source finding
CRITICIZES
Aonan Guan -> Copilot Agent
Aonan Guan demonstrated prompt injection vulnerability hijacking Copilot Agent via GitHub comments
Source finding

Source trail

Bloomberg via TechCrunch OX Security The NVD listing Rapid7's April Patch Tuesday analysis Shannon Trend Micro's analysis CrabTrap Mozilla's blog BleepingComputer Visual Studio Blog GitHub Changelog GeekWire

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…