chore(kaizen): weekly research scan 2026-06-12

[philmerrell]

Summary

• External scan: AWS Bedrock/AgentCore (5 items — Fable 5 GA, Interactive Shells, Agent-EvalKit), Strands Agents (1.43.0 today — closes #2635), FastMCP (CVE-2026-48710 starlette), MCP spec (SEP-2624 Interceptors), agentic UI/UX, frontier models, agent harness patterns, opencode (v1.17.x), pricing.
• Internal audit: 27 commits (skills-mode PRs testing pipelines #1–7 + PR-1/2 all merged); 7 consecutive Nightly CI failures; version-pin lag table (strands 3 behind, bedrock-agentcore 5 behind, starlette CVE-affected).
• Top 5 ideas in the dated research doc and queued in docs/kaizen/review-queue.md.

Headline findings

1. Strands 1.43.0 shipped today — closes the tracked #2635 count_tokens/toolResult=0 bug; the [2026-06-05] keystone target advances from 1.42 → 1.43.
2. Claude Fable 5 GA (claude-fable-5, $10/$50/M) — naming pattern shift breaks claude-opus-4.* string matching.
3. Nightly CI: 7 consecutive failures — prerequisite to investigate before landing dep bumps.
4. CVE-2026-48710 — our starlette==1.0.0 is in the affected range; 1-line fix.
5. AgentCore code interpreter security research (BeyondTrust) — read before enabling code-exec surface.

Review

• Read the research doc.
• Comment on the PR with reactions and any weekend POC findings — these become first-class signal for next Friday's kaizen-review-prep.
• POC promising ideas over the weekend.

Decision

Ship the doc to develop. Ranking into decisions happens in the kaizen-review-prep PR opened later this morning. Action on individual ideas happens in separate PRs the following week.

🤖 Generated with Claude Code