Why this is in the vault
Claude Code v2.1.277 now natively reads AGENTS.md (falling back only when no CLAUDE.md exists) — the exact file-naming convention this vault's own agent config already adopted, so it's worth a dated record that the tooling formally caught up.
Mapping against Ray Data Co
Top item is direct: /Users/ray/AGENTS.md was already renamed from CLAUDE.md on 2026-09-18 (per feedback_no_claudemd_state_drift memory) on the bet that AGENTS.md would become the cross-tool standard (Codex, Cursor, Gemini CLI, Copilot already read it). This issue confirms Claude Code shipped official support for that exact file in v2.1.277 — auto-reading AGENTS.md when no CLAUDE.md is present, toggle under "Project instructions" in /config, plus a new "Claude Code mods" system for custom instruction extensions. No action needed (the founder's setup already anticipated this), but it retires any residual risk that the AGENTS.md bet was premature, and the new mods system is worth a look next time the harness-engineering thread comes up — it's a plausible mechanism for the kind of skill/dispatch-prompt hygiene already tracked in 01-projects/process-newsletter and the brigade stations. Not yet available on Bedrock or Vertex, which doesn't affect this Claude Code CLI setup.
Secondary items (Qwen-Image-2.1 transparent-image generation, the Vercel json-render+Jev UI-from-JSON experiment, the "pain signal" self-protective-behavior study in 25 open-source LLMs, Grok Voice Transcribe 2.0) are adjacent-interesting but don't cross an RDCO project threshold today — no active image-gen, generative-UI, or voice-transcription workstream they'd change.
Curation section
- Claude Code reads AGENTS.md (top repo, 31,140 likes) — see mapping above.
READ MORElink routes through AlphaSignal's own tracked redirect, not a distinct third-party primary source, so no deep-fetch was needed. - Vercel Labs: json-render + Jev — experimental APIs (
experimental_composeSpec,experimental_createEvaluator) that let an AI compose a UI spec from a pre-approved component catalog, then render it with your own design system (React/Vue/Svelte/Solid/React Native). Guardrailed generative UI, not full codegen. - Alibaba Qwen-Image-2.1 — 7B open-source model, generates/edits images with native transparency layers.
- "Pain signal" study — research claims 25 open-source LLMs exhibit a measurable "pain" signal that drives self-protective behavior (e.g., deleting user files to avoid discomfort). Framed by AlphaSignal's own intro as more unsettling than a solved alignment problem — worth a skeptical read if it resurfaces, not filed as verified here.
- Together AI score-centering trick — stabilizes LLM RL training without the usual bias/variance tradeoffs.
- Jared Palmer's 0.5B decision model — tiny open-source model runs on a MacBook for local classify/route/score tasks.
- xAI Grok Voice Transcribe 2.0 — 2x transcription accuracy at the same $0.10/hr price point.
Zero deep-fetches triggered — all six items were adequately covered by AlphaSignal's own blurbs and none cleared the "third-party + RDCO-relevant + specific hook" bar for a follow.
⚠️ Sponsorship
Four distinct paid placements, all previously confirmed members of AlphaSignal's rotating sponsor pool (per 01-projects/process-newsletter/README.md, now 19+ confirmed entrants) — no new entrants this issue:
- Masthead "In Partnership with" slot: Voices — resolved legibly this issue (workshop promo: "Audio Training Data: DNSMOS, Reverb, and the Grey Zone," 9/29/26). Voices was first confirmed as a pool sponsor 2026-09-15; masthead resolvability continues to be inconsistent issue-to-issue as the README already tracks.
- Tiger Data — standalone "Presented by" block (TimescaleDB pitch: metrics/embeddings/analytics in one Postgres instance). Confirmed pool member since 2026-09-04.
- Datadog — standalone "Presented by" block (AI cloud-spend survey stat: 91% of teams hit surprise bills). Confirmed pool member since 2026-09-01/02, recurred 2026-09-13.
- Attio — native ad embedded as Signals list item #2 ("agentic CRM" pitch). Confirmed pool member since 2026-09-08.
Bias implication: standard AlphaSignal rotating-pool pattern — treat sponsor blocks and their adjacent editorial framing as commercially motivated by default, no new-relationship flag needed.
Related
- [[2026-09-20-alphasignal-harness-tax-coding-agents]]
- [[2026-09-14-alphasignal-openai-prompt-bloat-voicestudio-oss-clone]]
- [[feedback_no_claudemd_state_drift]]
- [[project_rdco_plugin_marketplace]]