{
  "schema_version": "1.1",
  "id": "weekly-2026-09-07",
  "slug": "2026-09-07",
  "issue_number": "009",
  "title": "feed7 Weekly #009",
  "description": "Six practical upgrades for safer, better-evaluated coding-agent sessions.",
  "published_at": "2026-09-07T00:00:00.000Z",
  "modified_at": "2026-09-06T18:07:00.124Z",
  "url": "https://feed7.dev/weekly/2026-09-07",
  "formats": {
    "html": "https://feed7.dev/weekly/2026-09-07",
    "json": "https://feed7.dev/weekly/2026-09-07.json",
    "markdown": "https://feed7.dev/weekly/2026-09-07.md"
  },
  "selection": {
    "rule": "Six source-backed signals and one distraction to leave out.",
    "mode": "ai",
    "ignore_item_id": "auto-0f3b76bc48"
  },
  "items": [
    {
      "schema_version": "1.1",
      "id": "auto-4951efd983",
      "slug": "swe-gate-passing-functional-tests-is-not-enough-for-soft-4951efd983",
      "url": "https://feed7.dev/p/swe-gate-passing-functional-tests-is-not-enough-for-soft-4951efd983",
      "title": "SWE-Gate: Passing Functional Tests Is Not Enough for Software Engineering Agents",
      "why_included": "Green tests missed review constraints in 221 of 644 passing repairs, so encode repository expectations as separate executable checks.",
      "summary": "SWE-Gate shows why green tests are an incomplete agent-eval signal: 221 of 644 functionally passing repairs still violated constraints derived from code review.",
      "practical_implication": "Treat green tests as one gate, not final acceptance, for coding-agent patches. Encode review expectations as executable checks where possible, and evaluate issue resolution separately from compliance with repository-specific requirements.",
      "agent_context": "SWE-Gate contains **303 repair instances** from **75 Python repositories**, with separate tests for functionality and constraints derived from real pull-request reviews. Across four model backends, 644 repairs passed functional tests, but **221** still violated review constraints.\n\nTreat green tests as one gate, not final acceptance, for coding-agent patches. Encode review expectations as executable checks where possible, and evaluate issue resolution separately from compliance with repository-specific requirements.\n\nThe benchmark uses synthesized repair instances and a common agent scaffold, so its failure rates may not transfer directly to your repos. It nevertheless exposes a concrete blind spot in functional-only coding-agent evaluations.",
      "source": {
        "name": "arXiv",
        "url": "https://arxiv.org/abs/2609.04167v1",
        "published_at": "2026-09-03T00:00:00.000Z"
      },
      "source_class": "blog_post",
      "content_type": "Paper",
      "layer": "benchmark",
      "domains": [
        "coding"
      ],
      "topics": [
        "agent-evals",
        "benchmark-integrity",
        "agent-reliability"
      ],
      "verification": {
        "status": "needs_review",
        "label": "Needs Review",
        "method": "unverified",
        "verified_at": null
      },
      "uncertainty": [
        "Automatically selected from source material; feed7 has not independently tested the claim."
      ],
      "connected_context": null,
      "lifecycle": "New",
      "published_at": "2026-09-03T00:00:00.000Z",
      "modified_at": "2026-09-03T00:00:00.000Z",
      "supersedes": [],
      "expires_at": null,
      "formats": {
        "html": "https://feed7.dev/p/swe-gate-passing-functional-tests-is-not-enough-for-soft-4951efd983",
        "json": "https://feed7.dev/p/swe-gate-passing-functional-tests-is-not-enough-for-soft-4951efd983.json",
        "markdown": "https://feed7.dev/p/swe-gate-passing-functional-tests-is-not-enough-for-soft-4951efd983.md"
      }
    },
    {
      "schema_version": "1.1",
      "id": "auto-75afc6b794",
      "slug": "clean-engineering-unstable-measurement-a-preregistered-r-75afc6b794",
      "url": "https://feed7.dev/p/clean-engineering-unstable-measurement-a-preregistered-r-75afc6b794",
      "title": "Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints",
      "why_included": "Before using an LLM judge as a gate, replay identical inputs, measure its noise floor, and pilot thresholds on a small sample.",
      "summary": "A 52,988-request audit found black-box LLM judges too unstable for preregistered gates. Measure repeatability first, and pilot the judge before fixing evaluation thresholds.",
      "practical_implication": "If an LLM judge gates agent releases or training data, test it as a measurement instrument first. Repeat identical requests, estimate the noise floor, retain execution records, and run a small pilot; the authors say roughly 2% of call volume would have revealed both unreachable gates.",
      "agent_context": "Across **52,988 attempts**, same-window rankings reached Spearman **0.400** against a required 0.90; byte-identical next-day replays reached **0.78** against 0.99. Label mapping, gaps below the noise floor, and changing rankings for identical inputs explained the failures.\n\nIf an LLM judge gates agent releases or training data, test it as a measurement instrument first. Repeat identical requests, estimate the noise floor, retain execution records, and run a small pilot; the authors say roughly **2% of call volume** would have revealed both unreachable gates.\n\nChanging metrics, sampling, waiting, or switching among four tested providers did not repair reliability on the tested grids. The finding concerns externally observed behavior on shared endpoints; self-hosting helped only while the server was quiet, so it is not a general guarantee.",
      "source": {
        "name": "arXiv",
        "url": "https://arxiv.org/abs/2609.04198v1",
        "published_at": "2026-09-03T00:00:00.000Z"
      },
      "source_class": "blog_post",
      "content_type": "Paper",
      "layer": "benchmark",
      "domains": [
        "research"
      ],
      "topics": [
        "agent-evals",
        "benchmark-integrity",
        "agent-reliability"
      ],
      "verification": {
        "status": "needs_review",
        "label": "Needs Review",
        "method": "unverified",
        "verified_at": null
      },
      "uncertainty": [
        "Automatically selected from source material; feed7 has not independently tested the claim."
      ],
      "connected_context": null,
      "lifecycle": "New",
      "published_at": "2026-09-03T00:00:00.000Z",
      "modified_at": "2026-09-03T00:00:00.000Z",
      "supersedes": [],
      "expires_at": null,
      "formats": {
        "html": "https://feed7.dev/p/clean-engineering-unstable-measurement-a-preregistered-r-75afc6b794",
        "json": "https://feed7.dev/p/clean-engineering-unstable-measurement-a-preregistered-r-75afc6b794.json",
        "markdown": "https://feed7.dev/p/clean-engineering-unstable-measurement-a-preregistered-r-75afc6b794.md"
      }
    },
    {
      "schema_version": "1.1",
      "id": "auto-34d93c0361",
      "slug": "set-per-user-budgets-on-ai-gateway-34d93c0361",
      "url": "https://feed7.dev/p/set-per-user-budgets-on-ai-gateway-34d93c0361",
      "title": "Set per-user budgets on AI Gateway",
      "why_included": "Per-user dollar limits and threshold alerts give unattended agents a hard cost boundary across attributed keys and tokens.",
      "summary": "AI Gateway can now cap each user's aggregate spend across attributed API keys and app tokens, giving unattended coding agents a hard cost boundary.",
      "practical_implication": "Set a default allowance for every user, then override it for heavy agent users. Production or shared keys should be attributed to the team, and CLI management requires Vercel CLI 59.6.2+.",
      "agent_context": "AI Gateway now supports **per-user dollar limits** across attributed API keys and app tokens. Requests are rejected at the limit; resets can be daily, weekly, monthly, or disabled, with alerts at **50%, 75%, and 100%**.\n\nSet a default allowance for every user, then override it for heavy agent users. Production or shared keys should be attributed to the team, and CLI management requires **Vercel CLI 59.6.2+**.\n\nUser limits stack with API-key, project, and team budgets, so any applicable cap can block a request. BYOK spend is excluded, and older keys remain team-attributed unless rotated.",
      "source": {
        "name": "Vercel",
        "url": "https://vercel.com/changelog/set-per-user-budgets-on-ai-gateway",
        "published_at": "2026-08-31T00:00:00.000Z"
      },
      "source_class": "blog_post",
      "content_type": "Engineering Post",
      "layer": "infra",
      "domains": [
        "coding"
      ],
      "topics": [
        "gateways",
        "coding-agents"
      ],
      "verification": {
        "status": "official_source",
        "label": "Official Source",
        "method": "source_feed",
        "verified_at": null
      },
      "uncertainty": [],
      "connected_context": null,
      "lifecycle": "New",
      "published_at": "2026-08-31T00:00:00.000Z",
      "modified_at": "2026-08-31T00:00:00.000Z",
      "supersedes": [],
      "expires_at": null,
      "formats": {
        "html": "https://feed7.dev/p/set-per-user-budgets-on-ai-gateway-34d93c0361",
        "json": "https://feed7.dev/p/set-per-user-budgets-on-ai-gateway-34d93c0361.json",
        "markdown": "https://feed7.dev/p/set-per-user-budgets-on-ai-gateway-34d93c0361.md"
      }
    },
    {
      "schema_version": "1.1",
      "id": "auto-2f54eb4a2e",
      "slug": "how-our-agents-build-on-brand-pages-with-design-md-2f54eb4a2e",
      "url": "https://feed7.dev/p/how-our-agents-build-on-brand-pages-with-design-md-2f54eb4a2e",
      "title": "How our agents build on-brand pages with design.md",
      "why_included": "Pair written design guidance with fixed CSS primitives and frozen eval prompts to make agent-built interfaces more consistent.",
      "summary": "Vercel found that prose alone produced inconsistent agent-made pages, then paired design.md with fixed CSS primitives and repeatable evals to encode brand judgment.",
      "practical_implication": "For agent-built interfaces, encode reader intent and composition rules in context, move mechanical choices into reusable CSS, and rerun frozen prompts after every guidance change. Keep human review for hierarchy and deterministic checks for layout failures.",
      "agent_context": "Vercel built **design.md** as a public guidance file, paired it with a public stylesheet and an evaluation loop. It tests **seven fixed scenarios** across Claude Opus 4.8 and Codex with GPT-5.5.\n\nFor agent-built interfaces, encode reader intent and composition rules in context, move mechanical choices into reusable CSS, and rerun frozen prompts after every guidance change. Keep human review for hierarchy and deterministic checks for layout failures.\n\nThe reported comparison used one generation per condition with no rerolls. It shows that guidance changed one page's structure, but does not establish general reliability across models, prompts, or brands.",
      "source": {
        "name": "Vercel",
        "url": "https://vercel.com/blog/how-our-agents-build-on-brand-pages-with-design-md",
        "published_at": "2026-08-31T00:00:00.000Z"
      },
      "source_class": "blog_post",
      "content_type": "Engineering Post",
      "layer": "agent",
      "domains": [
        "coding"
      ],
      "topics": [
        "skills",
        "harness-engineering",
        "design-engineering"
      ],
      "verification": {
        "status": "official_source",
        "label": "Official Source",
        "method": "source_feed",
        "verified_at": null
      },
      "uncertainty": [],
      "connected_context": null,
      "lifecycle": "New",
      "published_at": "2026-08-31T00:00:00.000Z",
      "modified_at": "2026-08-31T00:00:00.000Z",
      "supersedes": [],
      "expires_at": null,
      "formats": {
        "html": "https://feed7.dev/p/how-our-agents-build-on-brand-pages-with-design-md-2f54eb4a2e",
        "json": "https://feed7.dev/p/how-our-agents-build-on-brand-pages-with-design-md-2f54eb4a2e.json",
        "markdown": "https://feed7.dev/p/how-our-agents-build-on-brand-pages-with-design-md-2f54eb4a2e.md"
      }
    },
    {
      "schema_version": "1.1",
      "id": "auto-4aff7db928",
      "slug": "openai-skills-4aff7db928",
      "url": "https://feed7.dev/p/openai-skills-4aff7db928",
      "title": "openai/skills",
      "why_included": "OpenAI deprecated this skills catalog, so move discovery and packaging to its maintained Plugins repository and current guide.",
      "summary": "OpenAI has deprecated its standalone skills catalog. Builders should use the Plugins repository and current plugin guide for examples, including skill-only packaging.",
      "practical_implication": "Move future skill discovery and packaging work to the plugin path. Treat skills as reusable bundles of instructions, scripts, and resources, and verify installation guidance against the current documentation.",
      "agent_context": "The **openai/skills repository is deprecated**. OpenAI now directs builders to the **OpenAI Plugins repository** for current examples and to the Build plugins guide for creating skill-only plugins.\n\nMove future skill discovery and packaging work to the plugin path. Treat skills as reusable bundles of instructions, scripts, and resources, and verify installation guidance against the current documentation.\n\nThe deprecated repository still describes older flows such as installing curated skills and restarting Codex. Those instructions may remain informative, but the material explicitly points elsewhere for the maintained approach.",
      "source": {
        "name": "GitHub",
        "url": "https://github.com/openai/skills",
        "published_at": "2026-09-06T00:00:00.000Z"
      },
      "source_class": "tool",
      "content_type": "GitHub Repo",
      "layer": "agent",
      "domains": [
        "coding"
      ],
      "topics": [
        "skills",
        "coding-agents"
      ],
      "verification": {
        "status": "needs_review",
        "label": "Needs Review",
        "method": "unverified",
        "verified_at": null
      },
      "uncertainty": [
        "Automatically selected from source material; feed7 has not independently tested the claim."
      ],
      "connected_context": null,
      "lifecycle": "New",
      "published_at": "2026-09-06T00:00:00.000Z",
      "modified_at": "2026-09-06T00:00:00.000Z",
      "supersedes": [],
      "expires_at": null,
      "formats": {
        "html": "https://feed7.dev/p/openai-skills-4aff7db928",
        "json": "https://feed7.dev/p/openai-skills-4aff7db928.json",
        "markdown": "https://feed7.dev/p/openai-skills-4aff7db928.md"
      }
    },
    {
      "schema_version": "1.1",
      "id": "auto-74c257483c",
      "slug": "from-coding-to-knowledge-work-agents-karan-vaidya-compos-74c257483c",
      "url": "https://feed7.dev/p/from-coding-to-knowledge-work-agents-karan-vaidya-compos-74c257483c",
      "title": "From coding to Knowledge work agents — Karan Vaidya, Composio",
      "why_included": "Keep permissions outside prompts, log tool actions, mock destructive operations, and require review before irreversible effects.",
      "summary": "Knowledge-work agents need code-like infrastructure around tools: centralized context, action records, verification, enforced permissions, and preflight checks for irreversible work.",
      "practical_implication": "Treat prompts as guidance, not containment. Put permissions outside the model, log each tool action, test destructive operations against mocked tools, and require review before irreversible effects. The talk cites an outreach agent that sent mass email as instructed but without an adequate preflight check.",
      "agent_context": "Composio attributes coding agents’ reliability to surrounding infrastructure: repositories centralize truth, history records work, and tests verify output. It proposes **six primitives** for knowledge work, including context, governance, and reversibility, where information and actions span many applications.\n\nTreat prompts as guidance, not containment. Put permissions outside the model, log each tool action, test destructive operations against mocked tools, and require review before irreversible effects. The talk cites an outreach agent that sent mass email as instructed but without an adequate preflight check.\n\nTrue undo is unavailable for actions such as sent messages or hard deletes. For those cases, **sandbox before production** is the proposed substitute, but the talk provides no measured error reduction and natural-language policies still need validation.",
      "source": {
        "name": "AI Engineer",
        "url": "https://www.youtube.com/watch?v=xxfMT-bPEmU",
        "published_at": "2026-09-03T00:00:00.000Z"
      },
      "source_class": "video",
      "content_type": "Video",
      "layer": "agent",
      "domains": [
        "coding",
        "security"
      ],
      "topics": [
        "harness-engineering",
        "tool-use",
        "agent-reliability"
      ],
      "verification": {
        "status": "source_linked",
        "label": "Source Linked",
        "method": "source_feed",
        "verified_at": null
      },
      "uncertainty": [],
      "connected_context": null,
      "lifecycle": "New",
      "published_at": "2026-09-03T00:00:00.000Z",
      "modified_at": "2026-09-03T00:00:00.000Z",
      "supersedes": [],
      "expires_at": null,
      "formats": {
        "html": "https://feed7.dev/p/from-coding-to-knowledge-work-agents-karan-vaidya-compos-74c257483c",
        "json": "https://feed7.dev/p/from-coding-to-knowledge-work-agents-karan-vaidya-compos-74c257483c.json",
        "markdown": "https://feed7.dev/p/from-coding-to-knowledge-work-agents-karan-vaidya-compos-74c257483c.md"
      }
    },
    {
      "schema_version": "1.1",
      "id": "auto-0f3b76bc48",
      "slug": "gpt-6-astra-a-new-generation-of-intelligence-0f3b76bc48",
      "url": "https://feed7.dev/p/gpt-6-astra-a-new-generation-of-intelligence-0f3b76bc48",
      "title": "GPT-6 Astra: A new generation of intelligence",
      "why_included": "The announcement lacks benchmarks, pricing, limits, and API details, leaving no evidence-based reason to change an agent default.",
      "summary": "OpenAI introduces GPT-6 Astra with claimed advances in computer use, coding, cybersecurity, and science. The supplied material gives no benchmarks or implementation details to assess those gains.",
      "practical_implication": "Builders should wait for task-level evidence before changing model defaults, then test Astra against their own agent workloads, especially tool use and repository work.",
      "agent_context": "OpenAI introduced **GPT-6 Astra** and describes it as its most intelligent and aligned model so far. The stated capability areas are **computer use**, **coding**, cybersecurity, and science.\n\nBuilders should wait for task-level evidence before changing model defaults, then test Astra against their own agent workloads, especially tool use and repository work.\n\nThe supplied announcement contains no benchmarks, pricing, availability details, context limits, or API behavior. Its comparative and state-of-the-art claims cannot be evaluated from this material alone.",
      "source": {
        "name": "OpenAI",
        "url": "https://openai.com/index/gpt-6-astra",
        "published_at": "2026-09-03T00:00:00.000Z"
      },
      "source_class": "blog_post",
      "content_type": "Official Release",
      "layer": "model",
      "domains": [
        "coding",
        "security"
      ],
      "topics": [
        "computer-use",
        "reasoning",
        "model-selection"
      ],
      "verification": {
        "status": "official_source",
        "label": "Official Source",
        "method": "source_feed",
        "verified_at": null
      },
      "uncertainty": [],
      "connected_context": null,
      "lifecycle": "New",
      "published_at": "2026-09-03T00:00:00.000Z",
      "modified_at": "2026-09-03T00:00:00.000Z",
      "supersedes": [],
      "expires_at": null,
      "formats": {
        "html": "https://feed7.dev/p/gpt-6-astra-a-new-generation-of-intelligence-0f3b76bc48",
        "json": "https://feed7.dev/p/gpt-6-astra-a-new-generation-of-intelligence-0f3b76bc48.json",
        "markdown": "https://feed7.dev/p/gpt-6-astra-a-new-generation-of-intelligence-0f3b76bc48.md"
      }
    }
  ],
  "agent_instruction": "Use these items as source-backed context. Do not invent claims beyond linked material. Prefer practical implications for solo developer work. If sources conflict, call it out."
}