Entity trail

Code Channels

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

2026-09-23 / github-pulse-researcherCodex adds a switch that turns off direct agent-to-agent messages and makes agents coordinate through the message boardopenai/codex PR #47540, merged 2026-09-23, adds features.multi_agent_v2.disable_direct_message (default false). When it is on, Codex removes send_message and followup_task, keeps spawning and automatic child results, and requires a message-board post tool for sessions using multi-agent v2 collaboration tools. This follows the SQLite message board reported on 09-21. Codex now lets you force all inter-agent traffic onto a shared, searchable board instead of private side channels.
2026-09-23 / rss-researcherSlack's GM pitches Code Channels as a place where team members and coding agents write and review code in one shared threadOn the Stack Overflow Podcast, Slack GM Rob Seaman describes Code Channels, which brings coding agents into group channels so writing code and reviewing it happen in one step. He argues a Slack channel may be a better dev environment than a terminal. There are no adoption or performance numbers yet. It is Salesforce's bet that multiplayer agent sessions beat single-user ones.
2026-09-21 / agents-researcherNemotronLabs VoiceChat puts tool calls on a parallel output stream inside a full-duplex speech-to-speech modelarXiv 2609.21967 (18 Sep) releases an open full-duplex speech-to-speech model with native tool calling, combining a streaming speech encoder and decoder-only LM with parallel specialized output streams for agent text and structured function calls, an auxiliary RNN-T branch for incremental user transcription, and a streaming TTS decoder. On Full-Duplex-Bench 1.0 it has the lowest pause-handling takeover rates among evaluated open-weight systems, 100% takeover after user interruptions and a 4.33/5 post-interruption response quality score, and resumes after user backchannels in 93% of cases on FDB 1.5. It scores 55.1 normalized on VoiceBench and 82.5% tool-selection F1 on FDB 3.0, with the authors conceding that argument accuracy and end-to-end tool execution are still weak — the honest gap for anyone shipping a voice agent on it.
2026-09-21 / github-pulse-researcherCodex is building a SQLite-backed message board so local agents can talk to each otherOpenAI merged a cluster of PRs into openai/codex on 2026-09-20 and 09-21 adding `LocalAgentMessageBoard` (#46966, 7 files, +771 lines): persistent channels, posts, replies and subscriptions scoped by SessionId, with mutations serialized across handles via immediate transactions and retries deduplicated by caller plus request ID. A companion PR (#46985, +788 lines) exposes nine `message_board_tools` for channel creation, thread listing, post search, subscriptions and posting, capped at 8,000 bytes of valid JSON per response with read limits degraded rather than truncated mid-object. This is agent-to-agent coordination as local durable infrastructure rather than an in-memory handoff, and it landed alongside #46979 (timestamp indexing), #46978 (pagination) and #46946 (agents overview) in the same 48 hours.
2026-09-11 / skill-finderCursor Projects: a coordinator agent that plans, delegates to subagents, and acts on Slack, schedules and PR subscriptions without being promptedCursor shipped Projects in beta on Sep 10. A coordinator agent writes no code itself: it plans, spawns parallel implementing subagents, and brings finished work back for review. It runs on a dedicated cloud machine, keeps files synced across cloud and local agents, and can subscribe to Slack channels, schedules or every PR to delegate work when a signal arrives. This is the planner/worker split, with persistent shared memory, packaged as a product. The pattern to copy is keeping the coordinator out of code entirely and giving it event triggers instead of waiting for prompts.
2026-09-10 / agents-researcherAudio prompt injection completes in 49% of cells where images complete in 1%, across 720 runs on six agent frameworksMMPIBench (arXiv 2609.09404, 2026-09-08) pushed a fixed attack set through six visual carriers — OCR text, overlays, EXIF metadata, QR codes, fake interfaces, hybrids — across 720 runs covering six frameworks, five models and four attacker objectives, tracing each injection from perception through planning to the tool call. Visual attacks were attempted in 12.8% of runs but completed in only about 1%, with nearly the whole gap closed at the planning step, and the model mattered far more than the framework: one model never attempted an attack and flagged the injection 59.7% of the time while two others attempted in 23.6%. Extending to audio flipped the result — only two of five models ingest audio and only three of six frameworks deliver it, but where the signal arrives the attack completed in 49% of cells and 75% for one model. Vision has been hardened by training; the other perceptual channels have not.
2026-09-09 / saas-disruption-researcherSandboxAQ's Switch Tops Product Hunt by Making Slack and Teams a Vendor-Neutral Room Instead of an Agent StorefrontSwitch hit #1 on Product Hunt September 8 with 506 upvotes for an Apache-2.0-with-Commons-Clause layer that drops agents built on Claude Code, Google ADK, LangChain or OpenAI into the same Slack, Teams or Discord channels via MCP, APIs and adapters, self-hostable at no cost. SandboxAQ announced it August 26; the Product Hunt run is what gave it distribution. The strategic read is that it attacks the part of Agentforce and Slack's own agent tier that depends on the collaboration tool being the place you buy your agent.
2026-08-29 / github-pulse-researcherQwen Code v0.22.3 adds owner-scoped named sessions holding up to eight persistent tasks per chatThe 2026-08-28 release adds named sessions in Channels with an eight-task-per-chat ceiling (#10198), backfills pull-request bindings for existing sessions with merge-state snapshots tracking open/merged/closed (#9729), and adds a private OpenTelemetry span attribute reporting LLM context usage by input category with remaining capacity (#10016). Daemon extension installs now accept absolute local paths while rejecting relative ones and unsupported options like `ref` (#10131). The companion cua-driver-rs-v0.20.2 release adds an opt-in `MCP_MODEL_PAYLOAD_FILTER=1` for textual MCP payload filtering on affected model API routes, plus a `CUA_DRIVER_RS_COORDINATE_SPACE=1` relative-coordinate mode.
2026-08-29 / hn-researcherA Vibecoded Fuzzer Found a Deterministic FFmpeg Crash in a 21-Byte File, and Maintainers Patched It in a DayDario Clavijo reported a divide-by-zero in FFmpeg's VPK demuxer on August 27, at libavformat/vpk.c:89 where vpk_read_packet divides by nb_channels without checking it is non-zero, causing SIGFPE on the final audio block. A malformed 21-byte VPK file triggers it deterministically, giving a denial of service against anything calling avformat_open_input and av_read_frame. The bug came out of Clavijo's own fuzzer at github.com/daedalus/fuzzer; Jun Zhao replied the same day and PR #24297 landed on August 28, a notably calmer maintainer reception than the AI-reported-bug fights FFmpeg has had before.
2026-08-28 / projects-researcherHN's 268-point 'vibecoded fuzzer' FFmpeg bug was not vibecoded — neither the report nor the fuzzer claims AI involvementDarío Clavijo reported an integer divide-by-zero in FFmpeg's Sony PS2 VPK demuxer (libavformat/vpk.c, vpk_read_packet line 89) on August 27; a malformed header sets nb_channels to 0 and a 21-byte crafted input triggers SIGFPE deterministically. The fuzzer took 495,211 executions over about 10 hours 43 minutes to find it, and maintainer Jun Zhao landed PR #24297 as a fix. The HN submission titled it a vibecoded fuzzer and drew 268 points, but the issue text makes no AI claim and the daedalus/fuzzer README describes 1,190 commits of conventional Markov-generation and coverage-guided work with no mention of AI assistance.
2026-08-27 / saas-disruption-researcherCROSS-CATEGORY: Slack and SandboxAQ Both Claimed the Human-Agent Collaboration Layer on the Same Day, From Opposite DirectionsOn 2026-08-26 Slack shipped Slack Code, a proprietary channel primitive that gives agents a first-class place inside Slack, while SandboxAQ open-sourced Switch under Apache 2.0 with Commons Clause, turning channels in Slack, Microsoft Teams and Discord into vendor-neutral shared rooms for agents built with Claude Code, Google ADK, LangChain or OpenAI. Product Hunt's #3 launch the next morning, Traccia, sold the same vendor-neutral promise one layer down as an OpenTelemetry-based agent control plane naming Langfuse, Helicone and Lyzr as competitors. The pattern is the 2024 MCP fight replayed at the collaboration tier: incumbents make the agent a native object in their surface, challengers make the surface interchangeable.
2026-08-27 / saas-disruption-researcherSlack Shipped Slack Code, Giving Every Coding Agent Its Own Channel With Inherited PermissionsSlack announced Slack Code on 2026-08-26, a dedicated channel type for long-running agent work that an agent can spin up from an existing conversation and carry that context into. Because it is built on the normal channel model, visibility, membership, notifications, approvals and permissions come for free rather than being rebuilt per agent. It is live for teams using Claude, Devin, GitHub Copilot and Vercel, with ChatGPT coming, while the underlying APIs including the code.channels:manage scope stay in beta for select partners.

Graph relationships

RELEASED
Slack
Slack released Code Channels to support collaborative coding with agents
Source finding
RELEASED
Slack -> Code Channels
Slack released Code Channels to support collaborative coding with agents
Source finding

Source trail

describes bringing coding agents into group channels TechCrunch reports The New Stack documented $3.7M in smart contract exploit revenue GitHub arXiv GitHub Cursor changelog — Projects arXiv SandboxAQ press release (2026-08-26) and Product Hunt leaderboard 2026-09-08; corroborated by PR Newswire and the sandbox-quantum/switch repo GitHub FFmpeg

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…