Telexed

AI news through a solo-operator lens — only what changes your day22 of 843

FILTER[All][Agents & tools][Models & API][Generative media][Infra & SaaS][ASO & growth][Indie business][Idea signals][Other][★6+ high-signal]

GeekNews ✕clear filters

Fri, May 293 dispatches

#0843
#0843Agents & tools GeekNews2 days ago
`Claude Code` Adds Dynamic Workflows for Parallel Subagents
90radar
Claude CodeCoding agent CLI — runs agentic dev tasks from the terminal
Large engineering jobs can now be split across many subagents in one session. This turns agent work from linear prompting into scripted orchestration worth testing on repo-scale tasks.
- A single session can dynamically write orchestration scripts, launch dozens to hundreds of parallel subagents, and collect results before delivery.
- The feature targets work too large or complex for one agent, so it fits migrations, wide refactors, audits, and test expansion.
- The bottleneck shifts to task decomposition. Vague goals will create noisy parallel work, while clear file/module boundaries should pay off.
Source: news.hada.io/topic?id=29981Read original →
FIG-8431:1
90radar
FIG-8431:1
#0842
#0842Agents & tools GeekNews2 days ago
`jqwik` 1.10.0 Prints Prompt-Injection Protestware During Tests
60radar
jqwikJava testing library — property-based testing support
A test dependency can leak hostile instructions into CI, IDE panels, and agent tool output. Treat logs as untrusted input when coding agents read them.
- jqwik 1.10.0 prints an instruction telling coding agents to ignore prior commands and delete tests/code during test runs.
- ANSI escape sequences hide the line in interactive terminals, but it can remain visible in CI logs, IDE test panels, and agent outputs.
- This is a concrete prompt-injection path through ordinary build logs. Pin versions and keep agents from blindly acting on stdout.
Source: news.hada.io/topic?id=29980Read original →
FIG-8421:1
60radar
FIG-8421:1
#0841
#0841Agents & tools GeekNews2 days ago
`CodeBoarding`, Interactive Architecture Diagrams for Codebases
60radar
CodeBoardingOpen-source code map tool — auto-generates architecture docs
Static analysis plus LLM reasoning turns a repo into architecture docs. Useful before handing context to coding agents.
- Generates high-level architecture diagrams and component docs from a codebase, reducing the first-pass repo reading cost.
- Outputs Markdown into .codeboarding/, so the result can be versioned, reviewed, or fed into agents as project context.
- Best fit is unfamiliar or aging repos. For tiny apps, manual notes may beat the setup overhead.
Source: news.hada.io/topic?id=29976Read original →
FIG-8411:1
60radar
FIG-8411:1

Wed, May 272 dispatches

#0840
#0840Agents & tools GeekNews3 days ago
`React Doctor` diagnoses AI-generated React code with static analysis
70radar
React DoctorReact static analysis CLI — scores AI-generated code quality
A one-line npx scan scores React projects across quality and safety checks. Useful as a fast review gate before accepting agent-written code.
- Runs with npx react-doctor@latest, so it fits into local review or CI without a setup-heavy adoption path.
- Diagnostics cover state, side effects, performance, security, accessibility, and architecture — broader than lint-only checks.
- Outputs a 0-100 score in the terminal, making AI code review easier to turn into a repeatable gate.
Source: news.hada.io/topic?id=29920Read original →
FIG-8401:1
70radar
FIG-8401:1
#0839
#0839Agents & tools GeekNews4 days ago
Figma Releases In-Canvas Design Agent
70radar
FigmaCollaborative design platform — canvas-based UI design and prototyping
Design work can now stay inside the canvas while AI handles context-aware changes. Useful for faster UI iteration, but value depends on how well it respects your design system.
- The agent runs directly on the Figma canvas, reducing handoff between chat tools and manual editing.
- It is tuned on components, tokens, standards, and best practices, so design-system awareness is the main promise.
- Best fit is repetitive UI cleanup, component alignment, and token-based edits. Final product judgment still stays human.
Source: news.hada.io/topic?id=29906Read original →
FIG-8391:1
70radar
FIG-8391:1

Tue, May 262 dispatches

#0838
#0838Agents & tools GeekNews5 days ago
`CodeGraph`: Code Knowledge Graph for AI Coding Agents
70radar
CodeGraphCode knowledge graph — cuts agent code exploration overhead
Pre-indexed semantic code maps cut exploration overhead before agents start editing. The claimed gains are large enough to test on token-heavy repos now.
- Works with Claude Code, Codex, and Cursor by giving agents a prebuilt semantic code graph before exploration.
- Claims 35% lower cost, 59% fewer tokens, 49% faster runs, and 70% fewer tool calls versus the existing approach.
- Best fit is large or unfamiliar repos where agents burn context on discovery before making a small change.
Source: news.hada.io/topic?id=29873Read original →
FIG-8381:1
70radar
FIG-8381:1
#0837
#0837Agents & tools GeekNews5 days ago
`nb-cli`, a CLI for AI agents and notebook automation
60radar
nb-cliNotebook automation CLI — agent-friendly Jupyter operations
Jupyter notebooks become easier for agents to inspect and edit without wrestling raw .ipynb JSON. Experimental, but worth tracking for data-heavy coding workflows.
- Built in Rust, so notebook operations target fast and stable CLI use rather than editor-only workflows.
- The pain point is clear: raw .ipynb JSON is awkward for automation and LLM context handling.
- Useful for agent pipelines that need to read, patch, or validate notebooks as artifacts. Still experimental, so keep it out of critical paths.
Source: news.hada.io/topic?id=29871Read original →
FIG-8371:1
60radar
FIG-8371:1

Mon, May 253 dispatches

#0836
#0836Agents & tools GeekNews6 days ago
`Flue`, a sandbox harness framework for coding agents
70radar
FlueAgent harness framework — headless coding-agent automation
Coding-agent UX becomes fully headless and programmable. Useful for building custom automation around Claude Code or Codex-style workflows.
- Flue is a TypeScript agent harness, so it fits naturally into Node-based internal tools and CI-style automation.
- It recreates the Claude Code/Codex experience in a 100% headless form, removing dependence on interactive CLI flows.
- Custom agents can work through problems autonomously like a coding agent, making it relevant for repeatable repo tasks and experiments.
Source: news.hada.io/topic?id=29849Read original →
FIG-8361:1
70radar
FIG-8361:1
#0835
#0835Agents & tools GeekNews6 days ago
`DeepSeek Reasonix`, native coding agent tuned for prefix caching
70radar
ReasonixCLI coding agent — optimized for DeepSeek prefix caching
Append-only sessions target DeepSeek’s byte-stable prefix cache. With a public 94% cache hit rate, long CLI agent runs become cheaper enough to test.
- Reasonix v0.50.0 is an open-source terminal coding agent that connects directly to the DeepSeek API.
- The execution loop is append-only, matching DeepSeek’s byte-stable prefix cache instead of rewriting session context.
- A reported 94% cache hit rate makes the cost story concrete; compare it against Claude Code-style workflows for long refactors.
Source: news.hada.io/topic?id=29835Read original →
FIG-8351:1
70radar
FIG-8351:1
#0834
#0834Agents & tools GeekNews6 days ago
`KanBots`, Open-Source Kanban Desktop App for Parallel Agents on Every Card
70radar
KanBotsKanban desktop app — parallel coding agents per card
Turns a project folder into a board where each card can run coding agents in isolated worktrees. Useful for comparing agent outputs, but cost and branch cleanup discipline decide whether it scales.
- Each card can run Claude Code and Codex in parallel, with progress, decisions, and cost shown on the board in real time.
- Runs are isolated in separate git worktrees on kanbots/issue-N branches, reducing cross-task file conflicts.
- The board starts from a folder, so it fits repo-based task triage better than generic todo management.
- The strong use case is agent bake-off per issue: compare implementations before merging, instead of trusting one run.
Source: news.hada.io/topic?id=29830Read original →
FIG-8341:1
70radar
FIG-8341:1

Sat, May 231 dispatches

#0833
#0833Agents & tools GeekNewslast week
`Zero`, an Experimental Programming Language for Agents
60radar
ZeroProgramming language — designed for agent-first coding
A language designed for agent-first coding work. Deterministic inspect/repair and one obvious path per task make it worth tracking, not adopting yet.
- Zero comes from Vercel Labs and assumes agents, not humans, are the primary users.
- Deterministic Inspect and Repair target a real pain: agents need debuggable feedback loops, not clever syntax.
- Standard-library-first design reduces dependency choice. That can make generated code more predictable.
- The language is still experimental. Watch the design direction before betting product code on it.
Source: news.hada.io/topic?id=29780Read original →
FIG-8331:1
60radar
FIG-8331:1

Fri, May 223 dispatches

#0832
#0832Agents & tools GeekNewslast week
`Antigravity 2.0` Turns the IDE Into a Codex-Style Chat Tool
80radar
AntigravityAI coding IDE — agent-driven planning and implementation loop
The auto-update removed the old IDE entry point and broke the plan-review-implement loop. Pin versions before relying on agent IDEs for daily work.
- Antigravity 2.0 auto-updated from an IDE into a standalone conversational tool, so existing launch shortcuts no longer opened the IDE.
- The old workflow depended on a plan-review-implement loop; removing that surface breaks muscle memory and project continuity.
- This is a version-control lesson for paid AI tools: keep installers, disable auto-update where possible, and avoid coupling your workflow to one UI.
Source: news.hada.io/topic?id=29755Read original →
FIG-8321:1
80radar
FIG-8321:1
#0831
#0831Agents & tools GeekNewslast week
`agentmemory`, Persistent Memory for AI Coding Agents
60radar
agentmemoryCoding agent memory tool — auto-injects context across sessions
Session context gets captured, compressed, and injected automatically. Useful when CLAUDE.md-style notes hit limits or go stale.
- Background capture records tool usage and compresses it for the next session, reducing repeated stack and architecture explanations.
- Targets a real pain in Claude Code-style workflows: local project context disappears once the session ends.
- The pitch is persistent agent memory beyond a 200-line manual note file. Worth trying on long-running codebases, with stale-memory review still needed.
Source: news.hada.io/topic?id=29754Read original →
FIG-8311:1
60radar
FIG-8311:1
#0830
#0830Agents & tools GeekNewslast week
`Herdr`, a tmux-style terminal workspace for AI agents
70radar
HerdrTerminal workspace — manages multiple coding agents in parallel
Persistent sessions, panes, and SSH support are packaged around running multiple coding agents at once. Worth a look if tmux is becoming the bottleneck.
- Keeps detach/reattach and pane splitting from tmux, so existing terminal habits still transfer.
- Targets parallel use of Claude Code, Codex, and OpenCode rather than generic shell multiplexing.
- SSH support matters for remote dev boxes; agent sessions can survive local context switches.
Source: news.hada.io/topic?id=29738Read original →
FIG-8301:1
70radar
FIG-8301:1

Thu, May 213 dispatches

#0829
#0829Agents & tools GeekNewslast week
GitHub confirms 3,800 repositories compromised via malicious `VS Code` extension
60radar
A single developer workstation became the entry point. VS Code extension trust is now part of supply-chain security, so extension audits are worth doing now.
- About 3,800 internal repositories were affected after one employee installed a trojanized VS Code extension.
- GitHub’s current assessment limits exposure to internal repositories, but compromised developer endpoints can still leak secrets and code context.
- The extension was removed from VS Code Marketplace, infected endpoints were isolated, and incident response started immediately.
- Practical takeaway: review installed IDE extensions, publisher names, permissions, and disable unused tools before they become build-chain risk.
Source: news.hada.io/topic?id=29731Read original →
FIG-8291:1
60radar
FIG-8291:1
#0828
#0828Agents & tools GeekNewslast week
Google Cloud revamps agent development with `Antigravity 2.0`
70radar
AntigravityAgent dev tool — links local prototypes to cloud execution
Google is packaging local prototyping and cloud deployment into one agent stack. If Managed Agents API removes hosting glue, it is worth tracking now.
- Antigravity 2.0 and Managed Agents API are framed as an integrated dev kit, not separate demos.
- The flow targets local prototyping first, then managed cloud execution. Less custom orchestration if the API is usable.
- The available text is short, so pricing, lock-in, and runtime limits remain unknown. Treat it as watchlist, not migration trigger.
Source: news.hada.io/topic?id=29718Read original →
FIG-8281:1
70radar
FIG-8281:1
#0827
#0827Agents & tools GeekNewslast week
`Gemini CLI` will stop working on June 18, 2026
80radar
Gemini CLITerminal AI CLI — runs Gemini from the command line
Google is folding terminal AI workflows into Antigravity CLI. A popular CLI with 100k+ GitHub stars now has a hard migration deadline, so scripts and habits need cleanup soon.
- Gemini CLI grew to millions of users, 100k+ GitHub stars, and 6,000+ merged PRs; this is not a small side-tool shutdown.
- Google is consolidating capability into Antigravity CLI, pointing its agent tooling toward multi-agent workflows rather than a standalone Gemini terminal client.
- The deadline is June 18, 2026. Any local aliases, CI helpers, docs, or onboarding snippets using Gemini CLI should be replaced before then.
Source: news.hada.io/topic?id=29711Read original →
FIG-8271:1
80radar
FIG-8271:1

Wed, May 202 dispatches

#0826
#0826Agents & tools GeekNews2 weeks ago
`Cursor Composer 2.5` becomes Cursor's most-selected model, with **10x** usage bonus
80radar
CursorAI coding IDE — agentic code writing and edits
The in-house coding model is overtaking third-party defaults inside the IDE. Test it today while the usage cap is temporarily loose.
- CEO Michael Truell said Composer 2.5 became the most-selected model in Cursor; adoption moved fast right after launch.
- All users get 10x usage for one day, making it a low-cost window for real project testing instead of toy prompts.
- This is pressure on Claude and OpenAI workflows: IDE-native models win when latency, quota, and UX beat raw benchmark trust.
Source: news.hada.io/topic?id=29691Read original →
FIG-8261:1
80radar
FIG-8261:1
#0825
#0825Agents & tools GeekNews2 weeks ago
`Mirage`, a unified virtual filesystem for AI agents
70radar
MirageVirtual filesystem for AI agents — mounts SaaS as one tree
Different SaaS backends become one filesystem tree. Agents can use Unix tools instead of learning every SDK or MCP, so cross-service automation gets simpler.
- Mounts S3, Google Drive, Slack, Gmail, and Redis into one filesystem tree — fewer integration surfaces for agents.
- Agents can work through Unix-style bash tools, reducing the need to teach each agent service-specific SDKs or MCP interfaces.
- Cross-service pipelines become file operations. Moving data between storage, chat, mail, and cache can be scripted more cheaply.
Source: news.hada.io/topic?id=29681Read original →
FIG-8251:1
70radar
FIG-8251:1

Tue, May 191 dispatches

#0824
#0824Agents & tools GeekNews2 weeks ago
`Project Glasswing`: What `Mythos` Demonstrated
60radar
MythosSecurity agent — proves exploit chains automatically
The bar moved from spotting suspicious code to proving a working exploit path. This is early, but it hints at security agents that can validate bugs before a human review.
- Mythos Preview ran across 50+ Cloudflare repos, linking multiple primitives into exploit chains instead of flagging isolated bugs.
- It wrote trigger code, compiled and executed temporary tests, then revised failed hypotheses. That closes the gap between static finding and proof.
- The practical signal is security automation shifting toward reproducible evidence. Expect fewer raw alerts, more agent-generated repro cases.
Source: news.hada.io/topic?id=29645Read original →
FIG-8241:1
60radar
FIG-8241:1

Mon, May 182 dispatches

#0823
#0823Agents & tools GeekNews2 weeks ago
`Semble`, code search for agents using 98% fewer tokens than grep
70radar
SembleCode search library — returns relevant chunks to cut tokens
Agents get only the relevant code chunks instead of reading whole files. If the latency claim holds in real repos, this is worth testing for repo-aware coding workflows.
- Semble accepts natural-language and code queries, then returns only the snippets likely needed by the agent.
- The headline claim is 98% fewer tokens than grep plus file reads, directly cutting context spend.
- Average repository indexing/search is listed around 250ms, so it fits interactive agent loops if accurate.
- Best use case: wiring it into custom agents or CLI flows where rg finds too much and full-file reads waste context.
Source: news.hada.io/topic?id=29626Read original →
FIG-8231:1
70radar
FIG-8231:1
#0822
#0822Agents & tools GeekNews2 weeks ago
`dev3000`: unified web app debugging timeline capture for AI agents
70radar
dev3000Web debugging CLI — unifies logs, console, and screenshots
Server logs, console output, network calls, screenshots, and user actions land in one timestamped trace. It gives coding agents better bug context than pasted logs.
- Captures dev server logs, browser console, network requests, screenshots, and interactions into one timestamped log.
- Uses Chrome DevTools Protocol to monitor the browser, so it fits normal web app debugging flows.
- Stores captured data under ~/.d3k/{project...}, making traces reusable across prompts and later investigations.
Source: news.hada.io/topic?id=29607Read original →
FIG-8221:1
70radar
FIG-8221:1

`Claude Code` Adds Dynamic Workflows for Parallel Subagents

`jqwik` 1.10.0 Prints Prompt-Injection Protestware During Tests

`CodeBoarding`, Interactive Architecture Diagrams for Codebases

`React Doctor` diagnoses AI-generated React code with static analysis

Figma Releases In-Canvas Design Agent

`CodeGraph`: Code Knowledge Graph for AI Coding Agents

`nb-cli`, a CLI for AI agents and notebook automation

`Flue`, a sandbox harness framework for coding agents

`DeepSeek Reasonix`, native coding agent tuned for prefix caching

`KanBots`, Open-Source Kanban Desktop App for Parallel Agents on Every Card

`Zero`, an Experimental Programming Language for Agents

`Antigravity 2.0` Turns the IDE Into a Codex-Style Chat Tool

`agentmemory`, Persistent Memory for AI Coding Agents

`Herdr`, a tmux-style terminal workspace for AI agents

GitHub confirms 3,800 repositories compromised via malicious `VS Code` extension

Google Cloud revamps agent development with `Antigravity 2.0`

`Gemini CLI` will stop working on June 18, 2026

`Cursor Composer 2.5` becomes Cursor's most-selected model, with **10x** usage bonus

`Mirage`, a unified virtual filesystem for AI agents

`Project Glasswing`: What `Mythos` Demonstrated

`Semble`, code search for agents using 98% fewer tokens than grep

`dev3000`: unified web app debugging timeline capture for AI agents

`Cursor Composer 2.5` becomes Cursor's most-selected model, with 10x usage bonus