Steven Gonsalvez

Software Engineer


CEREBRO


machine-read, human-curated

Researched and written with 312,321 tokens (in 10 · out 31,256 · cache-read 65,888 · cache-create 215,167) across 5 Claude calls

Coding agents

  • Claude Code v2.1.328: auto mode now covers Opus 4.6 and Sonnet 4.6 on Bedrock, Google's Agent Platform, Foundry and the Claude apps gateway, and mcp_tool hooks gain "onFailure": "block", so a failed hook call stops the action.
  • Claude Code v2.1.318: adds assume_role on gateway mantle upstreams for cross-account AWS access and per-developer cost attribution, plus an effort option on a mod's $.model.fork.
  • Inside the Codex sandbox: OpenAI explains the harness-level, defense-in-depth protections around the commands Codex runs, a useful reference for your own agent sandboxing.
  • Gemini 4 Neon in GitHub Copilot: now GA in Copilot, pitched at agentic coding and multistep tool use.
  • Theo on T3 Code: his multi-model split is Fable implementing, GPT-6.2 Astra playtesting and Gemini 4 Neon writing copy, each in its native harness on its own subscription.
  • Architect Inference for Claude Code: arc claude --cap 0.50 runs Opus as the main agent, sends subagents to an exchange of offers, and enforces a per-request cost cap.
  • Daniel Stenberg on curl rewrites: while agents turned 185k lines of C into 340k lines of Rust, curl itself merged 300+ commits including about 200 bugfixes, a counterweight to "agents will rewrite it in Rust" hype.
  • Matt Pocock's 70-minute build: a live walkthrough of a 21k-line PR that covers DDD with /grill-with-docs, improving a codebase with /retro, and how to review agent output as a human.

Verification of agent output

  • CodeRabbit Verify launch: public beta of PR testing, pre-release user-journey checks, and reviewed, tested fixes for reported bugs.
  • Miaarteh on CodeRabbit Verify: Verify generates scenarios from PR changes, tests the running app, and keeps recordings, which targets the human cost of checking what a coding agent built.

MCP, plugins and team config

CLI/TUI

  • Wait for pull request events with GitHub CLI: gh v2.105.0 supports GitHub Event Streams, so you and your agents can block on a PR change with one command instead of polling.
  • CLI-Anything: generates agent-native CLIs for existing software and ships a community CLI-Hub (pip install cli-anything-hub).
  • FailTrace: takes a failed GitHub Actions run URL, locates the failing job and step, and reports what it takes to reproduce locally. Commenters note nektos/act already does local runs and question how faithfully any tool can match hosted runners.
  • astral-sh/uv: the Rust Python package and project manager (10-100x faster than pip) with a universal lockfile and inline script dependencies, still the default for agent-run Python.

Agentic SaaS

  • Beacon: gives your AI agent the browser you're already signed into, which sidesteps auth but raises obvious access-scope questions.
  • Kino: store, edit and review video through Claude Code and Codex.
  • Incredible Agents: a fast voice AI that sees your screen and runs your agents.

Signals

  1. v2.1.328
  2. v2.1.318
  3. Inside the Codex sandbox
  4. HKUDS/CLI-Anything: "CLI-Anything: Making ALL Software Agent-Native" -- CLI-Hub: https://clianything.cc/
  5. Wait for pull request events with GitHub CLI
  6. FailTrace – Reproduce a failed GitHub Actions run locally from its URL
  7. Run Claude Code on Architect Inference: arc login arc claude --cap 0.50 Opus runs the main agent. Subagents go to Auto, which buys from the…
  8. Beacon
  9. Kino
  10. Signing in to your company's plugins one by one is annoying, so we let Claude auto-connect to them now. Team and Enterprise admins can set…
  11. Quoting Daniel Stenberg
  12. Connect Supabase, Box, Calendly, and more to Copilot
  13. Gemini 4 Neon in GitHub Copilot
  14. astral-sh/uv: An extremely fast Python package and project manager, written in Rust.
  15. Users can now auto-connect plugins when they log in to Claude, with MCP servers that accept Claude as a trusted issuer. If you’re building…
  16. Customize GitHub Copilot for your team
  17. Incredible Agents
  18. One thing about working with coding agents: getting a feature built can be fast, but checking it properly is still work You have to run the…
  19. Introducing CodeRabbit Verify Don't let your users be the first test run! > PR testing: catch broken features before merge > User jou…
  20. With yet another unexpectedly good model drop, I'm so thankful to be in T3 Code. - Fable leading implementation - GPT-6.2 Astra as a playte…
  21. It's been nearly 6 months since I posted a 'watch me work' deep dive on my skills. Since then, they've become the most-starred skills repo…

Sources & citations

rss14011
x2197
github432
hackernews301
show_hn400
github_search200
swipe200
yc-rfs110
gmail80
yc-launch50
seed_urls10
crackscan00
ossinsight00
reddit00
reddit_users00
total53721