Gather — 5 April 2026

Security & Vulnerabilities #

• RCE and API Token Exfiltration Through Claude Code Project Files (CVE-2025-59536 / CVE-2026-21852) (Check Point Research) — CVSS 8.7 code injection via .claude/settings.json hooks on untrusted directories; CVSS 5.3 info disclosure via ANTHROPIC_BASE_URL set before trust prompt. Supply-chain class vuln: project config files trusted as metadata, not code.
• Critical Vulnerability in Claude Code Emerges Days After Source Leak (SecurityWeek) — Timing analysis linking the March 31 sourcemap leak to rapid CVE disclosure.
• Anthropic leaked Claude Code source via debug sourcemap (The Register, also Bloomberg, Axios) — JavaScript sourcemap for v2.1.88 shipped to npm; researcher Chaofan Shou discovered and posted within hours.
• Anthropic Claude Code Leak Analysis (Zscaler ThreatLabz) — Security-research angle on what the source reveals.
• Anthropic’s Claude Code Security GA after 500+ vulnerabilities found (VentureBeat) — Claude Code Security product now available; positioned as AI-powered security review.

Quota & Rate-Limit Crisis #

• Anthropic admits Claude Code quotas running out too fast (The Register) — Anthropic acknowledges users hitting limits “way faster than expected”; fix is “top priority”.
• Your Claude Code Rate Limit Is Draining Fast. Here Is Why. (Roborhythms, Mar 2026) — Attributes drain to broken prompt caching: model reprocesses full conversation at each turn instead of reading cache. Onset: March 23, 2026.
• Developers Using Anthropic Claude Code Hit by Token Drain Crisis (DevOps.com) — Industry-impact framing; users report 60% of session limit consumed in 30 minutes of coding.
• Claude Code users say they’re hitting usage limits faster than normal (The New Stack) — Early reporting on the trend before Anthropic confirmed the bug.

SDK & Platform Updates #

• Claude Agent SDK (renamed from Claude Code SDK) (GitHub - anthropics) — SDK rename with migration guide. New features: structured outputs with JSON schema validation, betas option for 1M-context window, plugins field, MCP tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint), in-process MCP servers via Python decorators.
• Agent SDK Overview (Claude API Docs) — Canonical reference for the renamed SDK.
• Claude Code by Anthropic — Release Notes (Releasebot) — Aggregated changelog; notable: /powerup interactive lessons, resume/performance improvements, PowerShell permissions fixes.

Workflow Patterns #

• Building Claude Code with Boris Cherny (Pragmatic Engineer / Gergely Orosz) — Interview with the Head of Claude Code. Boris ships 20-30 PRs/day running 5 parallel instances across 5 terminal tabs (each a separate git checkout). Workflow: start in plan mode → iterate on plan → one-shot implementation.
• Head of Claude Code: What happens after coding is solved (Lenny’s Newsletter, Boris Cherny) — Forward-looking framing: Claude Code hit 4% of public GitHub commits; daily active users doubling monthly.
• How Boris Uses Claude Code (standalone site) — Boris’s own documented setup. “Surprisingly vanilla” — no heavy customisation.
• Agentic Workflows with Claude: Architecture Patterns (Medium - Reliable Data Engineering) — Worker-Critic architecture: every creative agent paired with dedicated critic; strict separation (critics never create, creators never self-score). Proposes AGENTS.md as agentic counterpart to CLAUDE.md.
• wshobson/agents — Multi-agent orchestration for Claude Code (GitHub) — Intelligent automation toolkit for coordinating agents.

Author Updates #

• Auto mode for Claude Code (Simon Willison, 2026-03-24) — Commentary on Claude Code’s new auto-accept mode and when it helps vs hurts.
• Simon Willison NICAR 2026 workshop: Coding agents for data analysis (simonwillison.net) — 3-hour session for data journalists. Ran Datasette serving a viz/ folder while Claude Code vibe-coded Leaflet heat maps directly into it.
• Claude Code Remote Control (Simon Willison) — Drive Claude Code sessions from outside the terminal.
• Using Playwright MCP with Claude Code (Simon Willison TIL) — Integrating microsoft/playwright-mcp for browser automation from Claude Code sessions.
• Claude Code Can Debug Low-level Cryptography (Filippo Valsorda) — Cryptographer uses Claude Code to debug constant-time primitives; substantive case study on Claude’s reasoning depth for security-critical code.

Cross-links #

• AI Impact on Society The source leak + quota crisis are trust-erosion events worth tracking as governance/transparency signals.
• Vibe Coding Approaches Boris Cherny’s 5-parallel-terminals workflow + “plan mode → one-shot” pattern are canonical vibe-coding practices.
• Open vs Closed AI Ecosystems Accidental source leak of a flagship closed-model product is a notable irony — worth watching for second-order effects on open-weights arguments.
• Data, IP & Training Rights Source-code leak raises questions about what else Anthropic’s tooling exposes; supply-chain vuln CVEs touch on trust in closed tools.

Meta-observations #

• Emerging theme: Security is now a first-class concern for Claude Code — both vulnerabilities (CVEs, hook abuse, config-file trust) and product positioning (Claude Code Security GA). Two months ago this was absent from the journal.
• Emerging pattern: AGENTS.md proposed as agentic counterpart to CLAUDE.md — watch whether this becomes convention or remains one author’s term.
• Emerging pattern: Worker-Critic adversarial pairing (critics never create, creators never self-score) — an architectural pattern distinct from evaluator-optimizer because of the strict role separation.
• Keyword suggestion: “claude code CVE” or “claude code security vulnerability” — security-incident reporting is a new recurring category.
• Keyword suggestion: “claude code quota” / “claude code rate limit” — operational/economic issues now get more coverage than technique posts some weeks.
• Author to watch: Filippo Valsorda (words.filippo.io) — serious cryptographer, rare high-signal case study on Claude Code in security-critical contexts.
• Author to watch: Gergely Orosz (newsletter.pragmaticengineer.com) — already in vibe-coding config; his Boris Cherny interview warrants adding here too.
• Source to watch: howborisusesclaudecode.com — dedicated site from Claude Code’s creator, no other aggregator covers it.
• Source to watch: lennysnewsletter.com — landed a substantive Boris Cherny interview; worth monitoring for more insider perspectives.
• Gap: swyx search returned no Claude Code 2026 content — watch_authors list may need pruning (swyx has been less active on this specifically). Consider replacing with Filippo Valsorda or Gergely Orosz.
• Noise pattern: “Claude Code Tips X” articles proliferate — Substack and Medium each surface 3-4 listicles per week. Strong filter needed: prefer named practitioners (Boris, Simon, Filippo, Gergely) over SEO-driven roundups.