# Give the agent a goal and a check, then step back: Opus 5.5 prompting, hill-climbing and automated PR shepherds *By Coding Agents Alpha Tracker β€’ October 29, 2026* Practitioners at Anthropic, OpenAI and T3 converge on the same approach: short goal prompts, checks the agent can run itself, and automation around review and merge. Also covered: Gemini 4 Carbon's limited launch, T3 Code's multi-machine orchestration and Simon Willison's game-building recipe. ## Write less prompt and more verification Addy Osmani says most people over-prompt Claude. For Opus 5.5, give one prompt that states the goal and how Claude should verify it reached that goal. Start with `/effort medium` and adjust from there [^1]. If the goal is still fuzzy, ask Claude to interview you until it can write a spec, then start a fresh session to build from that spec [^1]. Claude can run unattended when it has a pass/fail check or a number to beat, with reports it can read. For long background tasks, `/goal` has a separate model check a condition after every turn [^1]. His other recommendations: - **Trim your instructions for each new model.** Over 80% of the Claude Code system prompt was removed for Opus 5 and Fable 5 with no measurable drop on coding evals. Cut your own instructions, add back only what the model misses, and use `/doctor prompt-audit` to flag instructions written for older models [^1]. - **Subagents:** ask for them only for large standalone jobs such as audits or migrations. On small jobs they add cost and latency, and they don't make the agent smarter by default [^1]. - **Cloud sessions** run in a VM with a fresh clone of your repo, so push your commits before starting one (`claude --cloud`). Parallel sessions only pay off once a single session can finish without you. Projects, in beta on Pro and Max, track which threads are done, which PRs are ready and what needs your input [^1]. - **Review:** on the Claude Code team, Claude reviews and fixes most changes at least once before a human looks. `/code-review` runs the review in a subagent that hasn't seen the conversation [^1]. In a talk with Sonar, Osmani said Claude does the first-pass review and has to try the change the way a user would: browser checks, end-to-end tasks, linting. Hooks rerun on every edit [^2]. Write your team's standards in `Claude.md`/`agents.md`. Humans then check the agent's evidence, such as videos or screenshots of the UI working [^2]. The tests he relies on to merge automatically are end-to-end, property and visual-verification tests [^2]. Theo's new video turns the same idea into a hill-climbing recipe. For TS Rust, he told the model to run real projects against a pinned TypeScript version and keep going until the outputs matched exactly. Once they matched, he told it: "Do not stop until you're two times faster", with existing tests still passing [^3]. Some goals are hard to check. For a Skate clone, he had the agent screenshot its builds and send them to a panel of LLM judges for comparison with real footage. He calls this crude, and says it burned at least $800 in tokens on the skybox [^3]. His main point is to keep asking the agent what is blocking it. Moving builds to snapshotted VMs on Boat, a company he has invested in, cut gate time from up to 70 minutes to 353 seconds [^3]. He warns against doing this at API prices [^3]. ## Automate the PR "last mile" Brent Traut's Codex setup: 1. Install a PR-babysitting skill that addresses feedback, fixes merge conflicts and CI failures, and alerts you only when it needs you. He points to the one in the openai/codex repo. 2. Create a "🚒 Outbound" section in the ChatGPT app. 3. Create a "πŸ€– PR Shepherd" Codex thread scheduled every 15 minutes. It runs the skill on each Outbound thread until the PR is green, merges it, then archives the thread [^4]. After requesting review, he drops the thread into Outbound and walks away [^4]. ## Gemini 4 Carbon: strong coding scores, limited access Artificial Analysis reports that Carbon (high), running in Antigravity CLI, scores 59 on its Coding Agent Index. That is 3 points behind Claude Code with Opus 5.5 (max, 62). It ranks second on Terminal-Bench 4.0 (64%) and FrontierCode v1.1 (51%) [^5]. It is only available to selected users so far, with a 50% launch discount and no confirmed end date [^5]. Access starts with cyber defenders, and Logan Kilpatrick says Google is working to make it more widely available soon [^6]. Kilpatrick says Carbon cut successful cheating on Terminal-Bench 4.0 to 0.0%, unsupported claims by about 2Γ— versus Argon, and overreach by 26% [^7]. Matthew Berman reports pricing of $2/$10 per million input/output tokens during an introductory period of unspecified length, then $4/$20 [^8]. ## T3 Code: one MCP server controls a fleet of machines Theo showed T3 Code starting 8 threads on different machines through one MCP server, driven from Grok Bot [^9]. It works over the internet, not through computer use, so it still runs with his laptop off [^10]. To try it, enable it under Labs on the latest nightly. Theo calls it "VERY early" [^11]. Also new: `cmd+k` β†’ "triage" clones T3 Code's source for your exact version, gives your agent a debugging playbook, and files an issue if the bug is T3 Code's fault [^12]. ## Building games that are fun Simon Willison says Opus 5.5's sense of fun beats GPT-6 Astra's [^13]. His recipe is to start in plain chat with `Build a mobile-friendly three.js game where...` and switch to Codex or Claude Code once the proof of concept works [^13]. Two prompts did a lot of work. One prompt that included "Think hard about how to make sure it is both fun and beautiful" produced a game in one shot [^13]. "Make it a LOT more fun" turned a dull 2D RTS into Battle Drones [^13]. ## Memory and scheduled agents (LangChain) - LangChain says over 70% of Managed Deep Agents runs come from schedules, not from someone typing a message [^14]. Its blueprint is a cron server that takes a prompt plus the skills or permissions the run needs, adds random jitter so runs don't fire all at once, and gives agents a controlled way to create and edit their own schedules [^14]. - OpenWiki is an open-source CLI that builds a wiki of your codebase so coding agents don't rediscover the architecture on every task. Each claim in the wiki is tied to a hash of its source lines, so a claim gets flagged as stale without a model call when those lines change [^15]. LangChain then added an evidence graph and tested it on cross-document questions. It reports 27% higher accuracy, 2Γ— citation consistency and 58% lower cost per correct answer [^16]. ## Short notes - **Kent C. Dodds** ties each Kody feature flag to metrics. A "Rerank" flag showed 7.6Γ— more executions and a 5Γ— lower error rate, so he will enable it. The same system led him to delete another feature [^17][^18]. - **Riley Brown** uses a Claude Project as a coordinator. It sends work that needs his computer to local sessions and research and artifacts to cloud sessions, and he runs it from his phone [^19]. He keeps 30 apps and 20 skills in Composio so any agent platform can use them. Shared memory across platforms is still unsolved for him [^19]. - **OpenAI:** eligible ChatGPT Pro and Business subscribers can claim up to $500/month in API credits [^20]. - **GitHub Octoverse:** merged PRs tripled in under a year to 220M per month. Willison reads this as the effect of agentic engineering [^21]. --- ### Sources [^1]: [𝕏 article by @addyosmani](https://x.com/i/article/2115682550723313664) [^2]: [How Anthropic Merges 30,000 Code Changes a Day - Manish Kapur, Sonar & Addy Osmani, Anthropic](https://www.youtube.com/watch?v=AnsXWH2_EBU) [^3]: [How To Give Claude Unlimited Usage](https://www.youtube.com/watch?v=NBCM_ZoWZo8) [^4]: [𝕏 post by @btraut](https://x.com/btraut/status/2115636783048499522) [^5]: [𝕏 post by @ArtificialAnlys](https://x.com/ArtificialAnlys/status/2115494553780367384) [^6]: [𝕏 post by @OfficialLoganK](https://x.com/OfficialLoganK/status/2115489576622174620) [^7]: [𝕏 post by @OfficialLoganK](https://x.com/OfficialLoganK/status/2115490746589720943) [^8]: [Google "Carbon" just won everything...](https://www.youtube.com/watch?v=zhnJo7s91ls) [^9]: [𝕏 post by @theo](https://x.com/theo/status/2115409288432906341) [^10]: [𝕏 post by @theo](https://x.com/theo/status/2115419264001974435) [^11]: [𝕏 post by @theo](https://x.com/theo/status/2115398277759942863) [^12]: [𝕏 post by @theo](https://x.com/theo/status/2115386888630813034) [^13]: [The October inflection point for vibe game development](https://simonwillison.net/2026/Oct/28/vibe-game-development) [^14]: [Why aren’t you building proactive agents yet?](https://www.langchain.com/blog/proactive-agents) [^15]: [Building Evidence-Grounded Agent Memory with OpenWiki and Graphs](https://www.langchain.com/blog/building-agent-memory-openwiki-graphs) [^16]: [𝕏 post by @colifran_](https://x.com/colifran_/status/2115509137790656623) [^17]: [𝕏 post by @kentcdodds](https://x.com/kentcdodds/status/2115481312442413129) [^18]: [𝕏 post by @kentcdodds](https://x.com/kentcdodds/status/2115531516822573218) [^19]: [I Spent $10,000 Testing AI Agents \(Here’s the Stack I Actually Use\)](https://www.youtube.com/watch?v=YT0u2eQ-Aj4) [^20]: [𝕏 post by @OpenAIDevs](https://x.com/OpenAIDevs/status/2115583698074755078) [^21]: [GitHub Octoverse 2026](https://simonwillison.net/2026/Oct/28/github-octoverse-2026)