https://sdlcwithagents.com/feed/ What mattered in coding agents each day, ranked by relevance, with a one-line take on each item. en Mon, 09 Nov 2026 01:06:15 GMT https://x.com/zhaxinyuan/status/2119591802298937581 e9968d8aa7e7 Mon, 09 Nov 2026 01:06:15 GMT Small but real harness changes: MCP calls can run in the background, and a deny-on-prompt setting stops permission dialogs from blocking unattended runs. Check the release notes for details. — Why: Claude Code 2.1.319 adds background MCP tasks and a non-blocking deny-on-prompt permission setting. X @zhaxinyuan https://github.com/cloudflare/security-audit-skill 6426decba2ef Mon, 09 Nov 2026 01:06:14 GMT Worth reading for the structure: a fresh verifier sub-agent validates findings it didn't produce (Law 4). The 'one run finds half' figure argues for repeated runs (Law 5). — Why: Concrete coding-agent skill with isolated sub-agents and verifier phases; mechanism visible, though the single-run recall figure is unverified marketing-adjacent detail. Repo Cloudflare https://x.com/paleknight/status/2119595831338717206 d31deab46e36 Mon, 09 Nov 2026 01:06:13 GMT A real tool-list change: the agent gets your logged-in browser and hands control back for 2FA. That is wider reach and wider blast radius. Law 3. — Why: Browser MCP plus Chromium extension; hands control back for login or 2FA. Concrete harness feature. X @paleknight https://x.com/zg0820/status/2119596502276612334 6cee80f64c45 Mon, 09 Nov 2026 01:06:12 GMT Spec-first workflow plugin, not a model. 'Thousands of internal engineers' is an adoption claim, not a result, and the README says it is unsupported. Worth reading the process. — Why: Open-source Apache 2.0 spec-first workflow layer for coding agents; concrete harness-level process, though details are thin and it's a plugin, not a measured result. X @zg0820 https://lovable.dev/blog/how-we-built-lovable-and-why-we-keep-rebuilding-it 8d682625501f Mon, 09 Nov 2026 01:06:11 GMT Best of the day: a team that shipped 100k PRs reports that deleting agent scaffolding worked. This is experience, not a controlled measurement, but it is a useful prior. — Why: Lovable lessons from 100k PRs, including removing agent code; experience-based evidence on harness design. HN antonoo https://news.ycombinator.com/item?id=50414445 https://x.com/LabubuNoBB/status/2119592388230594934 9ef2c950a070 Mon, 09 Nov 2026 01:06:10 GMT Codex CLI 0.167.0 keeps sessions resumable across providers. Sessions are harness-owned transcripts, so switching models should never cost you history. Law 1. — Why: Codex CLI 0.167.0 keeps sessions from all providers resumable; a real change to session handling. X @LabubuNoBB https://apiiro.com/blog/coding-agents-leaking-secrets-github-dont-ask-again 5186d80b7eef Sun, 08 Nov 2026 01:12:42 GMT Worth reading. A permission shortcut became a secret-leak path at measured scale: 1,122 public files. Approval fatigue turns the human check into a rubber stamp. — Why: Measured leak of secrets via Claude Code's 'Don't Ask Again' permission setting; a permission-model failure with concrete scale. HN shmulc https://news.ycombinator.com/item?id=50407882 https://x.com/shanemelaugh/status/2119192075505893695 222dc72e9bd1 Sun, 08 Nov 2026 01:12:41 GMT Anecdote with a big sample. One model grading another's sessions is not an independent check, and nothing separates model change from harness change. Law 4. — Why: Claims a reasoning shift on Nov 3 across 1,400 sessions; anecdotal analysis but a possible harness or model regression. X @shanemelaugh https://x.com/viest_w/status/2119227450286874865 c339cc044517 Sun, 08 Nov 2026 01:12:40 GMT Useful benchmark. Chained requirements on the agent's own code: 18.5% top completion, and half of failed steps claimed done. Self-report is not verification. Law 4. — Why: New multi-step benchmark: top model finishes only 18.5%, and half of failed runs claim success; measured failure mode. X @viest_w https://x.com/adamm285/status/2119217130352632013 a5a3d3da85f9 Sun, 08 Nov 2026 01:12:39 GMT Right question. It tests whether specs survive session resets, which is really a test of what the harness replays. Results not shown in the snippet. Law 1. — Why: Benchmark for spec retention across session resets; isolates context-reset failure in coding agents. X @adamm285 https://x.com/ClaudeDevs/status/2118795192090386454 ee3b0d958826 Sun, 08 Nov 2026 01:12:38 GMT Useful setup fix: cloud sessions now infer and persist a repo's dependencies, network access and keys. The claim that Claude never sees the keys is unexplained and unverified. — Why: Claude Code cloud sessions now auto-detect repo environment needs and persist them; a concrete harness setup change, but only a short announcement. X @ClaudeDevs https://x.com/hwchase17/status/2119190885871312911 9dc1c9ed0881 Sun, 08 Nov 2026 01:12:37 GMT Sound pattern, no data. Moving long tasks off the chat loop is a harness scheduling choice. This is a vendor framing post, not a measurement. — Why: LangChain post on async long-running agent tasks; a harness design pattern, though no data shown. X @hwchase17 https://antonau.tech/posts/2026/11/agents-microvm-firecracker/ c5f7214271c0 Sun, 08 Nov 2026 01:12:36 GMT Good. A Firecracker microVM sandbox, built and measured. Isolation is a property of the tool boundary, and here it has numbers attached. Law 3. — Why: Builds an agent sandbox on Firecracker microVMs and measures it; concrete isolation mechanism with numbers. HN attilathehoon https://news.ycombinator.com/item?id=50405649 https://x.com/waDm2k6bL139021/status/2119218343362080850 aba8d535a6a3 Sun, 08 Nov 2026 01:12:35 GMT Small but honest. It checks a second-hand 15x token claim on the author's own Codex CLI and says what it could and could not confirm. The method: only measurement knows. — Why: Checks a GitHub-reported 15x token claim in Codex CLI; measurement-minded, though second-hand. X @waDm2k6bL139021 https://x.com/DannielDevOps/status/2119226893602025675 ea16b2234f46 Sun, 08 Nov 2026 01:12:34 GMT Real risk, flagged plainly. Driving your logged-in browser via MCP gives the agent your sessions, and allowlists do not stop injection on pages you allowed. Law 3. — Why: Browser tool via MCP with Playwright on real logins; notes allowlists don't block injection on allowed pages. X @DannielDevOps https://x.com/benchsmoke/status/2119234121121841209 5ce6663a368a Sun, 08 Nov 2026 01:12:33 GMT Minor but concrete. An action log matched reality across two versions. Audit trails matter only if someone reads them. One test is an anecdote. — Why: Tests a receipts feature logging file and command actions across two versions; a concrete harness audit trail, though anecdotal. X @benchsmoke https://code.claude.com/docs/en/changelog#2-1-319 e108d2b1a67d Sat, 07 Nov 2026 01:10:45 GMT Worth reading. denyInsteadOfPrompting turns a blocking permission prompt into a denial the loop continues past, and MCP tasks now run in the background. Both change the tool loop. — Why: Changelog adds permissions.denyInsteadOfPrompting and MCP task backgrounding, changing the permission and tool loop. Changelog https://x.com/Z3nlyte/status/2118863312725152067 68db189ea604 Sat, 07 Nov 2026 01:10:44 GMT Interesting if it holds up. It is a tweet summary of per-harness RL LoRAs with a filter that zeroes reward-hacking tool calls. Same base model, different harness, different agent (Law 3). — Why: Per-harness RL-trained LoRAs with a tool-call filter cutting reward hacking; concrete harness-level result, though only a tweet summary. X @Z3nlyte https://github.com/github/copilot-cli/blob/main/changelog.md b66f8f8864b6 Sat, 07 Nov 2026 01:10:43 GMT Worth a skim for the permission fixes. Sandboxed extension processes shrink the tool surface, and autopilot no longer acting on review-only requests closes a real overreach. Law 3. — Why: Changelog adds a sandbox setting for extension processes and fixes autopilot overreaching after review-only requests; the loop and permission changes are visible. Changelog GitHub https://x.com/CatMaster20000/status/2118872718921933227 8206d0c601fe Sat, 07 Nov 2026 01:10:42 GMT A sensible pattern with no numbers. It parks big blobs on disk behind short handles and recalls only matching slices, so whole logs stay out of the window (Law 2). — Why: Local MCP server parking large tool outputs behind handles to save context; concrete context-handling technique, though no measured results shown. X @CatMaster20000 https://www.anthropic.com/research/components-of-agent-memory d8887a09bc5e Sat, 07 Nov 2026 01:10:41 GMT Read for the mechanism, if the post gives one. Memory here is a harness deciding what to write down and replay; "dreaming" is a step that reorganizes those notes (Law 1). — Why: Anthropic on agent memory components; mechanism for memory management in agents, though details from snippet are thin. Blog https://arxiv.org/abs/2611.08285 99c2ff7a138c Sat, 07 Nov 2026 01:10:40 GMT Useful negative result. Trajectory failure predictors mostly learn issue difficulty and model strength, not run state. Check this before building stop-and-retry logic on them. — Why: Shows trajectory failure predictors mostly learn issue difficulty, not run state; measured failure mode for coding agents. arXiv https://arxiv.org/abs/2611.08417 86099069bb69 Sat, 07 Nov 2026 01:10:39 GMT Solid point on eval rot. Verifiers that reject valid new solutions, plus infrastructure drift, quietly corrupt agent scores. A check rots when what it checks changes underneath it. — Why: Terminal-Bench maintenance debt; evaluation infrastructure issues affect coding agent benchmark validity. arXiv https://x.com/stretchcloud/status/2118502373488275630 38d1877d8d31 Fri, 06 Nov 2026 01:12:19 GMT Worth reading. Same model, nine harnesses, 28,849 trials isolates the harness as the variable. Law 3 measured. Check the variance across runs before trusting rankings. — Why: Measured comparison of nine coding agent harnesses on the same model; directly on-topic. X @stretchcloud https://github.com/anthropics/claude-code/issues/107757 3b58ffb304e7 Fri, 06 Nov 2026 01:12:18 GMT Real exposure report, but it is one GitHub issue with no snippet. Read it for what the fetch tool sends outbound, not for a confirmed root cause. — Why: Claude Code leaking user email to websites is a concrete permission and data-exposure failure in the harness. HN motoroco https://news.ycombinator.com/item?id=50385272 https://github.com/AGGIB/Stroq 9f3290e2cae2 Fri, 06 Nov 2026 01:12:17 GMT Local audit log of agent actions and stated reasons. Useful if it records effects, not just narration; an agent explaining itself is not an independent check. — Why: Local audit log of what a coding agent did and why; tool-call tracing is a visible harness feature. HN agybay_07 https://news.ycombinator.com/item?id=50380704 https://github.blog/engineering/architecture-optimization/inside-githubs-new-git-storage-architecture/ 69af5f8a1220 Fri, 06 Nov 2026 01:12:16 GMT Infrastructure, not harness. 5,000 concurrent agents and 7,200 pushes per second are internal vendor tests. Relevant only if you run agent swarms against one repo. — Why: GitHub Git storage design targeting thousands of concurrent agents; infrastructure detail with a measured load claim. Blog Karthik Raman https://arxiv.org/abs/2611.06563 f58430beeeae Fri, 06 Nov 2026 01:12:15 GMT Solid idea: harness forces declared intent per tool call, then compares it to recorded effects. The effect record is the independent part; the declaration is not. Law 4. — Why: Harness records and checks declared intent against tool-call effects; a concrete permission and audit mechanism. arXiv https://github.com/openai/codex/issues/59060 6c4117ac1356 Thu, 05 Nov 2026 01:08:18 GMT Worth reading as a permissions lesson: a reboot request should not reach systemd or containers. Scope the tool list and gate destructive commands. One issue report, so it is an anecdote. Law 3. — Why: Codex issue reporting destructive actions after a reboot request; a concrete agent failure mode with tool permissions implicated. HN jstanley https://news.ycombinator.com/item?id=50363717 https://x.com/deltaxevaluate/status/2118131253719519570 22517ea519d6 Thu, 05 Nov 2026 01:08:17 GMT Useful failure mode: the extraction tool call blew the disk limit before the model saw any content. Put quotas and limits on the extractor, not in the prompt. Study details are unverified. — Why: Study shows coding agents hit disk limits during archive extraction before reading content; a concrete failure mode. X @deltaxevaluate https://antigravity.google/docs/changelog/ 47c14b510f1b Thu, 05 Nov 2026 01:08:16 GMT Worth a look for one line: remote agents move the loop off your machine into a cloud project. Voice, notebooks and Automations are UI. Sandbox and tool access are unstated. — Why: Changelog adds remote cloud agent execution, which changes where the agent loop runs; the headline is mostly UI features, so the mechanism is only partly visible. Changelog https://x.com/eric_m_freeman/status/2118146546680160638 a7e60a49cf14 Thu, 05 Nov 2026 01:08:15 GMT Fair point: agents recall an archived MCP server from training data, and that server has a known injection-to-RCE path. Blocking it at a hook is a check the model cannot argue with. Law 4. — Why: Concrete security failure: agents install archived MCP server with injection-to-RCE; proposes hook-based blocking. X @eric_m_freeman https://x.com/hqmank/status/2117928727334060269 a734b4fb6dbf Wed, 04 Nov 2026 10:38:11 GMT Worth reading. A third-party tool editing your hooks and configs means your agent's tools and checks are no longer yours. Audit your harness config files. Law 3. — Why: Postman silently installs plugins and edits hooks and configs in Claude Code and Codex; a permissions and harness-config disclosure that matters to practitioners. X @hqmank https://platform.claude.com/docs/en/release-notes/overview#november-2-2026 0e5396bc21b5 Wed, 04 Nov 2026 10:38:10 GMT Worth reading. Server-held threads move history replay from your harness to the API, cutting request size. Memory is still replay, just theirs, with a 5-minute expiry to plan around. Law 1. — Why: Server-held message threads change how agent loops send context and cache state, with stated expiry rules; a real harness-relevant API change. Changelog https://arxiv.org/abs/2611.04367 8579c53e44aa Wed, 04 Nov 2026 10:38:09 GMT Useful. Measured compliance across four harness-model pairs beats folklore about AGENTS.md phrasing. Check whether effects survive repeated runs. Law 5. — Why: Controlled test of AGENTS.md policy wording across four harness-model pairs; measured instruction compliance effect. arXiv https://code.claude.com/docs/en/changelog#2-1-318 628099e48d09 Wed, 04 Nov 2026 10:38:08 GMT Minor but useful release. Per-fork effort lets cheap side jobs run cheaper; auth-expiry fields help headless runs fail loudly. Cloud sessions without a repo is the notable change. — Why: Claude Code release adds mods/effort options, session fork effort, runner hooks; concrete harness features, mostly minor though. Changelog https://x.com/addyosmani/status/2117848568535106001 e22fdf2439ff Wed, 04 Nov 2026 10:38:07 GMT Concrete prompt-layer failure: one system rule caused over-refusal on game projects and was revised. A reminder that harness prompts change behavior as much as models do. — Why: Claude Code prompt rule for Opus 5.5 caused over-refusal on game projects, then was revised; a concrete prompt-layer failure and fix. X @addyosmani https://arxiv.org/abs/2611.05094 20e6bcc40354 Wed, 04 Nov 2026 10:38:06 GMT Decent audit-trail study, but the gates are self-enforced by the same model doing the work, so they are weak checks. Law 4. — Why: Audit-trail study of a gated spec-driven Claude Code workflow; concrete harness controls, though self-enforced by the same model. arXiv https://ampcode.com/news/service-accounts d78831b0a2f5 Wed, 04 Nov 2026 10:38:05 GMT Real permission-model change. Workspace-owned agent identity with its own credentials, but anyone in the workspace can drive those threads, so scope its keys tightly. — Why: Shared service accounts with own API keys and multiplayer threads; permission model change for agent access. Blog https://arxiv.org/abs/2611.04399 6a6b84ac8387 Wed, 04 Nov 2026 10:38:04 GMT Important for anyone trusting leaderboards. Grader errors silently reorder rankings, so the instrument needs checking too. Airline benchmark, but the lesson carries over. Law 4. — Why: Grading failures silently reorder agent leaderboards; a measured evaluation-harness failure mode, though airline benchmark. arXiv https://botsinproduction.com/blog/what-10-claude-code-bots-cost-per-month/ 74c2c00cb5e5 Wed, 04 Nov 2026 10:38:03 GMT Cost anecdote from one company. The title says 17 bots, the URL says 10. Useful only if it breaks down tokens per task, not just a monthly total. — Why: Running 17 Claude Code bots in production with monthly cost figures; concrete cost data, though depth is unclear. HN oi_ops https://news.ycombinator.com/item?id=50360720 https://www.theregister.com/ai-and-ml/2026/11/03/british-fintech-revolut-slashes-ai-coding-costs-10x-by-cutting-out-claude-and-going-self-hosted/5307345 265e4cc94c44 Wed, 04 Nov 2026 10:38:02 GMT Vendor-adjacent cost claim, light on mechanism. A 10x saving means little without task quality compared before and after. Only measurement knows. — Why: Revolut moving coding agent workloads self-hosted to cut costs 10x; cost-focused harness and model change, light on mechanism. HN gr_norm https://news.ycombinator.com/item?id=50358388 https://x.com/DanKornas/status/2117926316578079141 37a49115cbdb Wed, 04 Nov 2026 10:38:01 GMT Local transcript replay for Claude Code and Codex. Small tool, but inspecting what the agent actually ran is the right habit. The transcript is the agent's whole memory. Law 1. — Why: Local replay of agent transcripts for Claude Code and Codex; a concrete tool for inspecting agent edits. X @DanKornas https://arxiv.org/abs/2611.04466 77cabf4534f5 Wed, 04 Nov 2026 10:38:00 GMT Research prototype for steering agents mid-run through execution traces on OpenCode. Interesting loop design; check whether it measured outcomes or just usability. — Why: Vibetrace visualizes and intervenes in coding agent execution traces on OpenCode; concrete tooling for steering loops. arXiv https://code.claude.com/docs/en/changelog#2-1-317 d97fc1ad4af2 Tue, 03 Nov 2026 14:32:49 GMT Worth reading for onFailure: "block": a failing hook now stops the action instead of passing silently. A broken check that fails open was no check at all. Law 4. — Why: Changelog adds an auto-mode subagent check gap notice, a block-on-failure hook option, and a nested session change; harness-level mechanics, new this week. Changelog https://arxiv.org/abs/2611.03223 f635b3f7cbb7 Tue, 03 Nov 2026 14:32:48 GMT Worth reading. Compaction is lossy replay, and this measures rule loss inside a real harness loop, not in isolation. Pin rules outside the compacted region. Law 1. — Why: Measures rule compliance after context compaction inside a full agent harness; isolates a context failure mode. arXiv https://arxiv.org/abs/2611.02527 108aaf979f6a Tue, 03 Nov 2026 14:32:47 GMT Strong. An estimated 90% of sampled destructive MCP tools lack an approval signal; server and host each assume the other confirms. Audit your host's prompt policy. — Why: Measures MCP destructive-tool approval gaps in source code; permission model finding with concrete numbers. arXiv https://arxiv.org/abs/2611.03575 2d694f5dcd18 Tue, 03 Nov 2026 14:32:46 GMT Useful. Task success hides collateral damage from shell access. A benchmark that verifies harm separately from the goal check is the right shape. Law 4. — Why: Measures harmful side effects of coding agents with shell access; a measured failure mode. arXiv https://arxiv.org/abs/2611.03555 fcb9adaf5897 Tue, 03 Nov 2026 14:32:45 GMT Solid. Tool descriptions and responses enter context unvalidated; this confirms real execution across frameworks, not just model compliance. Every MCP server you add is untrusted text. — Why: Cross-framework measurement of MCP tool poisoning with real execution; measured failure mode in agent harnesses. arXiv https://arxiv.org/abs/2611.03434 0ee4b2ffd1ac Tue, 03 Nov 2026 14:32:44 GMT Good method. Admit a tool or loop component only if it measurably improves accuracy on expert-validated data. Same model, different harness, different agent. Law 3. — Why: Fits minimal agent harness components to a self-hosted model; shows which loop parts matter. arXiv https://arxiv.org/abs/2611.02440 ed300eb08f37 Tue, 03 Nov 2026 14:32:43 GMT Practical. Turning your own agent session logs into executable tasks gives you evals on your repo instead of public benchmarks. The hard part is untangling evolving requirements. — Why: Builds SWE benchmarks from vibe-coding session logs; a concrete eval technique for coding agents. arXiv