Telexed

AI news through a solo-operator lens — only what changes your day30 of 412

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

GitHub Trending Weekly ✕clear filters

Mon, May 184 dispatches

#0412
#0412Agents & tools GitHub Trending Weeklyyesterday
`react-doctor` scores React code quality from the CLI
80radar
react-doctorReact diagnostics CLI — outputs a 0-100 quality score
A single npx run returns a 0-100 score plus concrete diagnostics. Useful as a cheap CI guardrail for agent-written React.
- npx -y react-doctor@latest . scans Next.js, Vite, and React Native projects with framework-aware rules.
- Checks cover state/effects, performance, architecture, security, accessibility, and dead code. That maps well to common agent regressions.
- The included GitHub Action can comment on PRs and expose a score output. Set fail-on to block low-quality changes.
- react-doctor.config.json supports rule, file, and override ignores. Use narrow overrides to keep coverage intact.
- install writes React best-practice guidance into Claude Code, Cursor, Codex, OpenCode, and 50+ agents.
Source: github.com/millionco/react-doctorRead original →
FIG-4121:1
80radar
FIG-4121:1
#0411
#0411Agents & tools GitHub Trending Weeklyyesterday
`Bun`: one JavaScript toolchain for runtime, bundling, tests, and packages
80radar
BunJavaScript toolchain — runtime and package manager in one
One executable covers node, package install, scripts, tests, and TS/JSX execution. Fewer moving parts make small app stacks faster to set up and easier to keep lean.
- bun run index.tsx supports TypeScript and JSX out of the box, cutting the usual transpiler setup for quick prototypes.
- The CLI includes bun test, bun install, bunx, and script running, so one binary can replace several Node.js dev dependencies.
- Built in Zig on JavaScriptCore, it targets lower startup time and memory use than typical Node-based tooling.
- Linux, macOS, Windows, Docker, npm, and Homebrew installs are covered; adoption friction is low for existing projects.
Source: github.com/oven-sh/bunRead original →
FIG-4111:1
80radar
FIG-4111:1
#0410
#0410Agents & tools GitHub Trending Weekly2 days ago
`Superpowers`: skill framework for coding agents
60radar
SuperpowersCoding-agent skills framework — enforces spec-to-TDD workflow
A reusable skills layer turns coding agents into spec-plan-TDD executors. Useful if your agent often jumps into code too early.
- Runs on Claude Code, Codex CLI, Codex App, Cursor, Copilot CLI, and others — not tied to one agent harness.
- Workflow is explicit: clarify intent, produce readable spec chunks, get design sign-off, then generate an implementation plan.
- The method pushes red/green TDD, YAGNI, and DRY; its value is guardrails, not a stronger base model.
- Subagent-driven execution can run for hours when the plan is tight. Best fit: scoped refactors, feature slices, test expansion.
Source: github.com/obra/superpowersRead original →
FIG-4101:1
60radar
FIG-4101:1
#0409
#0409Agents & tools GitHub Trending Weekly2 days ago
`Pyrefly`: Meta’s Fast Python Type Checker and Language Server
80radar
PyreflyPython type checker — fast checks plus IDE features
Python projects get fast type checks plus IDE features in one tool. Strong fit for typed backends and AI-heavy codebases where slow feedback kills iteration.
- Claims 1.85M+ lines/sec and checks PyTorch 15x faster than Mypy and Pyright; save-time IDE rechecks usually finish under 10ms.
- Runs as both CLI and language server, covering navigation, autocomplete, hover info, inlay hints, and semantic highlighting with consistent results.
- Built-in Pydantic and Django support reduces false friction for typical SaaS backends and API projects.
- Migration path is practical: pyrefly init, pyrefly suppress, and pyrefly infer let teams start with one file instead of a full rewrite.
Source: github.com/facebook/pyreflyRead original →
FIG-4091:1
80radar
FIG-4091:1

Thu, May 141 dispatches

#0408
#0408Other GitHub Trending Weekly6 days ago
`easy-vibe`: beginner course for shipping AI apps with SaaS and payments
50radar
easy-vibeCourse — from AI app building to SaaS and payments
Starts from plain-language prompting, then walks into full-stack shipping with Stripe, SaaS UI, and RAG visuals. Useful as a fast onboarding map; advanced builders will mostly mine it for teaching flow and idea-validation sections.
- The curriculum goes beyond prompting into a SaaS capstone, payment integration, multi-product UI, and WeChat Mini Program backend flows.
- Visual tutorials, simulated IDE guidance, and clickable RAG/terminal demos lower early friction better than text-only docs.
- A new appendix adds idea sourcing, JTBD, Double Diamond, and The Mom Test, so it covers validation as well as building.
- Best fit is onboarding non-engineers or juniors into AI product building; experienced builders will not get much technical depth.
Source: github.com/datawhalechina/easy-vibeRead original →
FIG-4081:1
50radar
FIG-4081:1

Wed, May 133 dispatches

#0407
#0407Agents & tools GitHub Trending Weekly7 days ago
`CloakBrowser`: stealth Chromium as a drop-in for Playwright/Puppeteer
50radar
CloakBrowserBrowser tool — patched Chromium for stealth automation
Uses a real Chromium build with source-level fingerprint patches instead of JS spoofing. Useful when browser automation gets blocked, but policy and abuse risk make it a selective tool, not a default stack choice.
- Claims 30/30 bot-detection tests passed, including Cloudflare Turnstile, FingerprintJS, and BrowserScan.
- Keeps the Playwright/Puppeteer workflow mostly intact: swap imports to launch() and keep the rest of the code nearly unchanged.
- Ships 49 C++ fingerprint patches across canvas, WebGL, audio, fonts, GPU, WebRTC, timing, and automation signals.
- humanize=True adds mouse, keyboard, and scroll behavior simulation; first run auto-downloads a ~200MB Chromium binary.
- Includes a self-hosted profile manager via docker run ... cloakbrowser-manager, positioning it against Multilogin-style setups.
Source: github.com/CloakHQ/CloakBrowserRead original →
FIG-4071:1
50radar
FIG-4071:1
#0406
#0406Agents & tools GitHub Trending Weekly7 days ago
`agentmemory`: persistent memory layer for coding agents
70radar
agentmemoryMemory layer — persistent context across agents
Adds a shared memory server across agent sessions, so setup choices and code context carry over automatically. Broad MCP/hook support plus strong retrieval benchmarks make it worth testing now.
- One server works across Claude Code, Cursor, Codex CLI, Gemini CLI, Cline and other MCP/HTTP clients, so context is no longer trapped in one tool.
- Retrieval claims are strong: 95.2% R@5 and 98.6% R@10 on LongMemEval-S, well above its BM25-only fallback.
- It stores implementation decisions like auth stack, file paths, and test coverage, then injects them into later sessions without manual re-explaining.
- v0.9.0 added agent-memory.dev, @agentmemory/fs-watcher, and tighter MCP/server syncing so hooks, viewer, and memory state stay aligned.
Source: github.com/rohitg00/agentmemoryRead original →
FIG-4061:1
70radar
FIG-4061:1
#0405
#0405Generative media GitHub Trending Weekly7 days ago
`SuperSplat`: browser-based editor for 3D Gaussian Splats
50radar
SuperSplat3D Gaussian Splat editor — browser-based edit, optimize, publish
A free open-source 3D Gaussian Splat editor runs fully in the browser, covering inspection, editing, optimization, and publishing. No install lowers the barrier fast, but it matters mainly if 3D capture or spatial media is already in your stack.
- Covers the full loop: inspect, edit, optimize, and publish 3D Gaussian Splats in one web app, which cuts tool switching.
- Runs in the browser with no install, so testing workflows or client-side demos is much lighter than desktop-only tools.
- Local dev is simple: Node.js 18+, npm install, then npm run develop and open localhost:3000. Easy to fork and customize.
- Localization is already structured with static/locales plus src/ui/localization.ts, useful if you want a white-label or multilingual tool.
Source: github.com/playcanvas/supersplatRead original →
FIG-4051:1
50radar
FIG-4051:1

Tue, May 121 dispatches

#0404
#0404Other GitHub Trending Weeklylast week
`TabPFN`: fast foundation model for tabular prediction
40radar
TabPFNModel — tabular ML without preprocessing
A strong shortcut for classification and regression on structured data without feature scaling or one-hot encoding. GPU-first and best under 100k rows / 2k features, so it is worth testing for scoring or forecasting before building a heavier pipeline.
- pip install tabpfn gets you classifier and regressor defaults, with checkpoint download on first fit; setup is light for experiments.
- Skip scaling and one-hot encoding entirely. That cuts prep work and makes it attractive for messy product or business datasets.
- Inference is GPU-oriented: ~8GB VRAM works, 16GB helps on larger sets; CPU is only practical around <=1000 samples.
- Prediction should be batched. Repeated single-row predict calls recompute the training set and can become almost 100x slower.
- The project also points to TabPFN Client for hosted inference and extensions for SHAP, feature selection, outlier detection, and synthetic data use cases.
Source: github.com/PriorLabs/TabPFNRead original →
FIG-4041:1
40radar
FIG-4041:1

Mon, May 115 dispatches

#0403
#0403Other GitHub Trending Weeklylast week
`PageIndex`: vectorless RAG with tree-based reasoning
50radar
PageIndexRAG tool — searches via document trees, no vector DB
Replaces embeddings and chunking with a document tree plus LLM-guided search. Interesting if vector recall keeps missing long-doc answers, but cost/latency need real-world validation.
- Retrieval is split into two steps: build a table-of-contents-style tree, then run reasoning-based tree search over it.
- The pitch is no vector DB and no chunking, with page/section references that make retrieval paths easier to inspect.
- It claims 98.7% on FinanceBench, so the strongest fit is document-heavy domains where relevance beats semantic similarity.
- Recent additions push it beyond a single file: file-system tree indexing, a chat product, and MCP/API access for agent workflows.
Source: github.com/VectifyAI/PageIndexRead original →
FIG-4031:1
50radar
FIG-4031:1
#0402
#0402Agents & tools GitHub Trending Weeklylast week
`addyosmani/agent-skills`: reusable engineering skills for coding agents
50radar
agent-skillsSkill set — reusable workflows for coding agents
Turns senior-engineering workflows into reusable SKILL.md packages and lifecycle commands like /spec, /build, and /ship. Useful if you want stricter agent output without building your own prompting layer, but it's process scaffolding, not a model capability jump.
- Maps the dev lifecycle into 7 commands: /spec, /plan, /build, /test, /review, /code-simplify, /ship.
- Packages review gates, testing habits, and implementation workflows as portable skills, so agent behavior stays more consistent across tools.
- Supports multiple agent shells such as Claude Code, Cursor, Gemini CLI, Windsurf, and OpenCode, which lowers lock-in.
- The practical value is in standardizing execution quality; it won't make a weak agent smarter, but it can reduce sloppy output and missed steps.
Source: github.com/addyosmani/agent-skillsRead original →
FIG-4021:1
50radar
FIG-4021:1
#0401
#0401Agents & tools GitHub Trending Weeklylast week
`Flue`: headless TypeScript agent harness framework
60radar
FlueFramework — runs UI-less agent workflows in code
A fully programmable agent runtime that keeps Claude Code-style workflows but drops the human-in-the-loop UI. Promising if you want deployable agents across Node.js, Cloudflare, or CI, but the experimental API makes it an early bet.
- Runs agents in a default virtual sandbox via just-bash instead of full containers, so latency and cost can be much lower for high-traffic workloads.
- Puts most agent behavior in Markdown such as skills, context, and AGENTS.md, which reduces orchestration code and makes prompt-layer changes faster.
- Targets multiple runtimes including Node.js, Cloudflare, GitHub Actions, and GitLab CI/CD; useful when the same agent must ship across app backend and automation jobs.
- Typed outputs are built into the flow with schema validation in session.prompt(), which helps when wiring agents into production endpoints.
- Support-agent setup mounts knowledge from an R2 bucket directly into the filesystem and uses built-in search tools, a practical pattern for docs-backed support bots.
Source: github.com/withastro/flueRead original →
FIG-4011:1
60radar
FIG-4011:1
#0400
#0400Agents & tools GitHub Trending Weeklylast week
`UI-TARS Desktop` Adds Free Remote Browser and Computer Operators
70radar
UI-TARS DesktopDesktop agent — free remote browser and computer control
Open multimodal agents are getting usable outside closed IDEs. Free remote browser and computer control makes this worth testing for UI automation and ops work.
- The stack ships two tracks: Agent TARS for CLI/Web UI workflows and UI-TARS Desktop for native GUI control across desktop and browser.
- The 2025-06-12 update added free Remote Computer Operator and Remote Browser Operator, lowering the setup cost for agent-driven automation.
- Agent TARS CLI v0.3.0 added multi-tool streaming, timing stats, and an event stream viewer, which helps when debugging long agent runs.
- This is broader than a model demo: local/remote operators plus MCP tool integration make it more practical for repetitive QA, back-office, and browser tasks.
Source: github.com/bytedance/UI-TARS-desktopRead original →
FIG-4001:1
70radar
FIG-4001:1
#0399
#0399Agents & tools GitHub Trending Weeklylast week
`DeepSeek-TUI`: Terminal Coding Agent with Auto Model Routing
70radar
DeepSeek-TUITerminal coding agent — all-in-one TUI with rollback
Brings a full coding-agent stack to the terminal with streamed reasoning, approval gates, and workspace rollback. Strong if you want Codex/Claude Code-style flow on DeepSeek without building your own wrapper.
- Runs from the deepseek command and supports file edits, shell, git, web, sub-agents, and MCP in one TUI.
- --model auto also picks thinking level per turn, lowering manual model switching during mixed-cost tasks.
- Ships with 1M-token context handling, prefix-cache cost telemetry, and session resume for long jobs.
- /restore and side-git snapshots add rollback without touching the repo's own .git, which is safer for active projects.
Source: github.com/Hmbown/DeepSeek-TUIRead original →
FIG-3991:1
70radar
FIG-3991:1

Sun, May 102 dispatches

#0398
#0398Agents & tools GitHub Trending Weeklylast week
`Ouroboros`: spec-first Agent OS for replayable AI coding
60radar
OuroborosAgent OS — spec-first AI coding workflow
Adds a local-first runtime that turns agent coding into a replayable, policy-bound workflow. Worth testing if you want less prompt thrash and more verifiable runs across multiple coding CLIs.
- Uses a fixed loop of interview -> crystallize -> execute -> evaluate -> evolve, pushing requirements capture ahead of code generation.
- Targets multiple runtimes including Codex CLI, Claude Code, OpenCode, Hermes, Copilot CLI, and Kiro, reducing lock-in.
- The pitch is a 3-stage automated evaluation gate plus an immutable seed spec, so architecture drift and manual QA get constrained early.
- Setup is oriented around local tooling and MCP registration, with ooo interview "..." as the entry point instead of freeform prompting.
Source: github.com/Q00/ouroborosRead original →
FIG-3981:1
60radar
FIG-3981:1
#0397
#0397Infra & SaaS GitHub Trending Weeklylast week
`InsForge`, open-source backend stack built for coding agents
60radar
InsForgeBackend stack — all-in-one backend layer for agents
Puts auth, Postgres, storage, functions, hosting, and an OpenAI-compatible gateway behind one agent-friendly control layer. Worth tracking if you want a self-hostable alternative to stitching together Supabase + Vercel + AI proxy logic.
- Agents can operate it through an MCP Server or CLI + Skills, so backend setup and debugging become callable tools instead of manual dashboard work.
- The product surface is broad: Authentication, Database, Storage, Model Gateway, Edge Functions, site deployment, plus private preview compute.
- The Model Gateway exposes an OpenAI-compatible API across multiple providers, which lowers switching cost for LLM-backed apps.
- Self-hosting starts with Docker Compose and Node.js, making local end-to-end backend testing possible before committing to a hosted stack.
Source: github.com/InsForge/InsForgeRead original →
FIG-3971:1
60radar
FIG-3971:1

Sat, May 91 dispatches

#0396
#0396Agents & tools GitHub Trending Weekly2 weeks ago
`Browserbase Skills` plugs Browserbase automation into `Claude Code`
60radar
Browserbase SkillsBrowser automation skill pack — unifies CLI, traces, cookie sync
A ready-made skill pack turns Claude Code into a browser automation agent with bb CLI, tracing, CAPTCHA handling, and UI testing. Useful if you want one install to cover QA, scraping, and auth-heavy browser tasks.
- The bundle ships 12 focused skills, including browser control, bb CLI workflows, tracing, anti-bot detection, cookie sync, and UI testing.
- Install flow is short: npx skills add browserbase/skills, then /plugin install browse@browserbase inside Claude Code.
- site-debugger and browser-trace stand out because they target flaky automation debugging, not just happy-path browsing.
- cookie-sync plus persistent Browserbase contexts makes logged-in site automation much more practical than stateless browser runs.
Source: github.com/browserbase/skillsRead original →
FIG-3961:1
60radar
FIG-3961:1

Fri, May 83 dispatches

#0395
#0395Generative media GitHub Trending Weekly2 weeks ago
`ACE-Step UI`: polished local frontend for `ACE-Step 1.5` music generation
50radar
ACE-Step UIMusic UI — streamlines local ACE-Step generation
A Spotify-like frontend makes local AI music generation far more usable than raw model tooling. If you already have GPU headroom, free and unlimited beats paying monthly for lightweight song prototyping.
- Pairs a polished UI with ACE-Step 1.5, covering full songs, instrumentals, lyrics, batch runs, and prompt reuse in one flow.
- Pushes the strongest local pitch: no subscription, no queue limits, 100% local, which matters when iterating on many variants.
- Advanced controls go beyond consumer music apps: reference audio, cover transforms, repainting, seeds, and inference-step tuning.
- The catch is hardware and setup friction. Value is highest for creators already comfortable running GPU-heavy tools locally.
Source: github.com/fspecii/ace-step-uiRead original →
FIG-3951:1
50radar
FIG-3951:1
#0394
#0394Agents & tools GitHub Trending Weekly2 weeks ago
`CocoIndex`: Incremental Context Engine for Long-Horizon Agents
60radar
CocoIndexContext engine — incremental indexing for fresh context
Keeps agent context continuously fresh by recomputing only changed data. Useful when your app ingests docs, code, or Slack and stale RAG is the bottleneck.
- Re-indexes only the delta, so changed files get re-embedded without rerunning the full pipeline.
- Targets codebases, PDFs, meeting notes, inboxes, Slack, and videos, which fits multi-source agent backends.
- The API is declarative Python with connectors like localfs and postgres; setup looks closer to ETL than agent orchestration.
- It also ships a coding-agent skill for v1 patterns, lowering the chance an AI IDE generates the wrong integration code.
Source: github.com/cocoindex-io/cocoindexRead original →
FIG-3941:1
60radar
FIG-3941:1
#0393
#0393Agents & tools GitHub Trending Weekly2 weeks ago
`OpenAI Symphony`: spec and reference impl for autonomous coding-run orchestration
50radar
OpenAI SymphonyCoding orchestration — connects task queues to PRs
Moves agent ops up a level: watch a work queue, spin isolated coding runs, and attach CI, review, and walkthrough artifacts before merge. Still an engineering preview, but the spec alone is useful if you want a homegrown task-to-PR pipeline.
- Targets work management, not live supervision: a board event can trigger isolated implementation runs that return CI status, PR feedback, complexity analysis, and videos.
- Two paths are offered: build from SPEC.md, or run the experimental Elixir reference in elixir/README.md. That lowers the barrier to cloning the pattern without waiting for a polished product.
- The repo explicitly says trusted environments only. This is not a drop-in autopilot for production repos yet.
- It assumes harness engineering is already in place, so the immediate value is highest for teams or solo operators who already enforce tests, CI, and review gates.
Source: github.com/openai/symphonyRead original →
FIG-3931:1
50radar
FIG-3931:1

Thu, May 710 dispatches

#0392
#0392Other GitHub Trending Weekly2 weeks ago
`DocuSeal`: Open-Source E-Signature Stack You Can Self-Host
60radar
DocuSealE-signature stack — self-hosted with embeds and API
A practical way to add signing flows without paying DocuSign first. Docker deploy, embeddable forms, and API/webhooks make it usable now for document-heavy SaaS.
- Ships a WYSIWYG PDF form builder with 12 field types, so non-technical ops work does not need a custom UI first.
- Supports multiple submitters, SMTP emails, storage on disk or S3/GCS/Azure, covering the boring infra pieces usually missed in v1.
- Provides API, webhooks, and embedded signing/form-builder options for React, Vue, Angular, or plain JS, which lowers integration cost.
- Default deploy is a single docker run with SQLite; DATABASE_URL lets you move to PostgreSQL or MySQL when the product hardens.
Source: github.com/docusealco/docusealRead original →
FIG-3921:1
60radar
FIG-3921:1
#0391
#0391Other GitHub Trending Weekly2 weeks ago
`Scrapling`: adaptive Python web scraping with anti-bot fetchers and scalable spiders
50radar
ScraplingPython scraping framework — adaptive selectors, anti-bot built-in
A single Python library now spans one-off scraping to concurrent crawls, with adaptive selectors, proxy rotation, and built-in anti-bot fetchers. Useful if you need unstable pages or catalog monitoring, but the real value depends on how often site layouts break.
- StealthyFetcher and DynamicFetcher target modern anti-bot flows like Cloudflare Turnstile, so setup friction is lower than stitching multiple scraping tools.
- Adaptive selection can re-locate saved elements after DOM changes, which cuts maintenance on price tracking, listing sync, and competitor monitoring jobs.
- The spider layer supports concurrent multi-session crawls with pause/resume, real-time stats, and streaming, covering the jump from script to crawler in one package.
- Feature surface is broad: fetchers, parsers, spiders, proxy rotation, CLI, and MCP; good for Python-heavy stacks, less relevant if scraping is only occasional.
Source: github.com/D4Vinci/ScraplingRead original →
FIG-3911:1
50radar
FIG-3911:1
#0390
#0390Agents & tools GitHub Trending Weekly2 weeks ago
`jcode`, a coding-agent harness built for multi-session workflows
60radar
jcodeCoding agent harness — lightweight for multi-session runs
Optimized for many concurrent agent sessions with unusually low memory and near-instant startup. Worth a look if you care more about local orchestration efficiency than another chat wrapper.
- With local embeddings off, one session uses 27.8 MB PSS; even the full mode is 167.1 MB, below several agent CLIs listed.
- At 10 sessions, baseline memory is 117.0 MB and full jcode is 260.8 MB`; rivals in the table climb to 1.6-3.2 GB**, which matters for parallel runs.
- Time to first frame is 14.0 ms; the comparison table puts Codex CLI at 882.8 ms and GitHub Copilot CLI at 1518.6 ms.
- The pitch is not model quality but harness efficiency: multi-session operation, customizability, and low overhead as the main product surface.
Source: github.com/1jehuang/jcodeRead original →
FIG-3901:1
60radar
FIG-3901:1
#0389
#0389Agents & tools GitHub Trending Weekly2 weeks ago
`awesome-codex-skills`: Practical `Codex` Skill Library
50radar
awesome-codex-skillsSkill library — reusable Codex workflows and automation
Installable skills add task-specific execution rules and external actions to Codex. Useful as a starter kit for repeatable CLI/API automation.
- Skills install into $CODEX_HOME/skills and activate by metadata, so setup stays lightweight and reusable.
- The repo covers coding, writing, collaboration, and analysis tasks, making it more than a prompt snippet list.
- Each skill is a folder with SKILL.md instructions, which keeps workflows versionable and easy to adapt.
- Examples like multi-agent orchestration and app actions push Codex beyond code generation into operational automation.
Source: github.com/ComposioHQ/awesome-codex-skillsRead original →
FIG-3891:1
50radar
FIG-3891:1
#0388
#0388Agents & tools GitHub Trending Weekly2 weeks ago
`Ruflo`: multi-agent orchestration layer for `Claude Code`
60radar
RufloOrchestration layer — adds swarms and MCP to Claude Code
Adds a full coordination layer on top of Claude Code: swarms, memory, hooks, federation, and an MCP server. Useful if you want agent workflows without building your own orchestrator, but the full install is much heavier than a plugin-only trial.
- The full npx ruflo init path exposes 98 agents, 60+ commands, 30 skills, hooks, a daemon, and MCP-backed memory.
- Plugin installs stay light with slash commands and agent definitions only, but skip MCP features like memory_store and swarm_init.
- Positioning is clear: let coding agents coordinate across machines, sessions, and trust boundaries instead of acting as isolated copilots.
- Workspace footprint differs sharply by install path; the full setup writes .claude/, .claude-flow/, CLAUDE.md, helpers, and settings.
Source: github.com/ruvnet/rufloRead original →
FIG-3881:1
60radar
FIG-3881:1
#0387
#0387Other GitHub Trending Weekly2 weeks ago
`Maigret`: username-based OSINT across **3,000+** sites
40radar
MaigretOSINT tool — large-scale username account tracing
A single username can fan out into accounts, profile data, and linked identities across 3,000+ sites with no API keys. Useful for founder research, competitor checks, and abuse screening, but it sits in a gray zone where privacy and false positives need handling.
- Default runs hit the top 500 sites; -a expands to the full database, and --tags narrows by country or category for faster targeted checks.
- It can extract profile details, follow linked accounts, and recursively search discovered usernames or IDs, so one lead often turns into a wider identity graph.
- There is a Python-importable library plus Docker and web UI options, which makes it easy to plug into internal review flows or lightweight ops tooling.
- --ai can turn raw findings into a short summary through an OpenAI-compatible API, but the real value is structured collection rather than the model layer.
Source: github.com/soxoj/maigretRead original →
FIG-3871:1
40radar
FIG-3871:1
#0386
#0386Agents & tools GitHub Trending Weekly2 weeks ago
`mattpocock/skills`, Composable Agent Skills Instead of Heavy Process Kits
50radar
mattpocock/skillsAgent skills pack — modular prompt-based workflows
Break agent workflows into small reusable skills, not rigid process layers. Easy to copy for prompt assets and task setup.
- Install is one command: npx skills@latest add mattpocock/skills, then run /setup-matt-pocock-skills for tracker, labels, and docs path.
- The design argues against GSD, BMAD, and Spec-Kit style process ownership; smaller prompt units are easier to debug and swap.
- /grill-me and /grill-with-docs target the biggest agent failure mode: bad requirement capture before code starts.
- The repo pushes shared language and lightweight conventions so agents waste fewer words and pick up project jargon faster.
Source: github.com/mattpocock/skillsRead original →
FIG-3861:1
50radar
FIG-3861:1
#0385
#0385Idea signals GitHub Trending Weekly2 weeks ago
`Dexter`: open-source autonomous agent for financial research
50radar
DexterOpen-source agent — finance research with live data
A vertical agent pairs planning, self-checking, and live filings instead of stopping at generic chat. Useful as a blueprint for domain agents, but direct adoption is niche unless you ship finance workflows.
- Breaks financial questions into structured tasks, executes tools, and self-validates before returning an answer.
- Uses live statements and market datasets, so the moat is domain data access plus agent control, not just another wrapper.
- Includes loop detection and step limits; that matters if you want unattended runs without runaway cost.
- Ships with evals via LangSmith and LLM-as-judge scoring, which is stronger than the usual demo-only agent repo.
Source: github.com/virattt/dexterRead original →
FIG-3851:1
50radar
FIG-3851:1
#0384
#0384Agents & tools GitHub Trending Weekly2 weeks ago
`Craft Agents` OSS: document-first multi-agent workspace with MCP/API setup
70radar
Craft AgentsAgent workspace — document-first multi-agent GUI
A GUI-first agent workspace aims to remove MCP/API setup friction by letting agents wire sources, skills, and credentials directly. Interesting if you want Claude Code-style power without living in the CLI.
- Uses the Claude Agent SDK and Pi SDK together, with a document-centric UI instead of a code-editor-first workflow.
- Claims source setup can be done in-chat: add Linear, Slack, paste MCP JSON, or hand over an OpenAPI spec and let the agent configure it.
- Local MCP support matters: stdio servers can run as subprocesses from npx, Python scripts, or local binaries without a restart loop.
- Apache 2.0 and “built with itself” make it a useful reference for anyone building an agent-native desktop product, not just using one.
Source: github.com/lukilabs/craft-agents-ossRead original →
FIG-3841:1
70radar
FIG-3841:1
#0383
#0383Agents & tools GitHub Trending Weekly2 weeks ago
`Zed`, a high-performance multiplayer code editor from the Atom and `Tree-sitter` team
40radar
ZedCode editor — built for speed and multiplayer coding
Built around speed and real-time collaboration, not just another local editor. Cross-platform availability is practical today, but no web client yet keeps it from being a full default for every workflow.
- Positioning is clear: a high-performance, multiplayer editor, which puts collaboration in the core product rather than as a plugin add-on.
- Install paths already cover macOS, Linux, and Windows via direct download or package managers, so trial cost is low across common dev setups.
- Web is still unavailable and tracked separately, which limits use on locked-down machines, quick handoff links, and browser-first workflows.
- The repo exposes build docs for all three desktop platforms, a useful signal that self-hosting, patching, or deep customization is realistic.
- License compliance is strict enough that CI can fail on dependency metadata, so contributors should expect more operational rigor than a casual side project.
Source: github.com/zed-industries/zedRead original →
FIG-3831:1
40radar
FIG-3831:1