Agent Infrastructure Becomes Its Own Category — Day 221

GitHub trending on September 21, 2026, four days since the last snapshot in this series, and the gap tells its own story: affaan-m/ECC has kept climbing to 263,813 stars, cloudflare/security-audit-skill has more than doubled from 7,380 to 18,110, and anthropics/financial-services debuts as Anthropic’s own entry into vertical-specific agent tooling. The bigger shift is structural. coder/coder, trycua/cua, BuilderIO/agent-native, and ECC itself are no longer four unrelated repos. They’re four answers to the same question: what does an agent run inside of. “Agent infrastructure” has become a trending category in its own right, not just a label applied after the fact.

Anthropic Ships Into Financial Services

anthropics/financial-services (35,403 stars, +260 today): a new repository from Anthropic itself, entering the trending page with no fanfare in its listed description.

This is the first Anthropic-authored repo in this series aimed at a specific industry vertical rather than a general capability. Where claude-code and knowledge-work-plugins are horizontal tools usable by anyone, a repo scoped to financial services signals the same pattern seen across the broader ecosystem all year: once general-purpose agent tooling stabilizes, the next wave specializes by industry. Anthropic building its own vertical entry, rather than leaving it to third parties, is worth noting on its own.

cloudflare/security-audit-skill More Than Doubles in Four Days

cloudflare/security-audit-skill (18,110 stars, +2,428 today): up from 7,380 stars on September 17. That’s +10,730 in four days, averaging roughly 2,680 stars a day, an acceleration rather than a slowdown from the +927 it posted on its first tracked day.

A multi-phase security audit skill, producing independently verified, machine-readable findings, going from a fresh entrant to nearly 20K stars inside a week says more about appetite for auditable, non-hallucinating security tooling than about Cloudflare’s brand alone. Machine-readable findings matter here specifically because they’re the difference between an agent skill that produces a report a human has to trust blind, and one that produces output another system can verify or act on directly.

security-audit-skill validates: verifiable output, not just capability, is what turns a security-adjacent agent skill into a fast riser.

The “Agent Infrastructure” Category Solidifies

Four repos on today’s board aren’t building agents. They’re building what agents run inside of:

• coder/coder (16,082 stars, +379 today): “Secure environments for developers and their agents,” explicitly naming agents as first-class occupants of a dev environment, not an afterthought bolted onto human tooling.
• trycua/cua (25,190 stars, +1,018 today): “computer-use 2.0,” open-source drivers, cross-OS fleets, and benchmarks for training, evaluation, and data generation around computer-use agents specifically.
• BuilderIO/agent-native (5,260 stars, +98 today): a framework for building agentic apps from the ground up, rather than adding agent features to an existing app architecture.
• affaan-m/ECC (263,813 stars, +826 today): still the largest repo in this series by a wide margin, and still growing, four days after debuting at 260,330.

None of these four existed as a distinct category eighteen months ago. Agent environments, agent-native app frameworks, and cross-agent optimization harnesses were, until recently, just “infrastructure” in the generic sense. Now they’re specific enough to have their own naming conventions and their own trending slots.

The category-formation pattern validates: once enough agents exist, the market stops asking what agents can do and starts asking what agents need to run well.

claude-code’s Slow, Steady Climb

anthropics/claude-code (147,168 stars, +419 today): up from 145,540 on September 17, a gain of 1,628 over four days, a noticeably calmer pace than the fast risers elsewhere on the board. For a repo already past 145K stars, steady low-hundreds daily growth looks less like stalling and more like a maturing base rate: the audience most likely to star it already has.

Two Signals From the Edges

mihail911/modern-software-dev-assignments (4,578 stars, +172 today): assignments for a Stanford course, “The Modern Software Dev” (Fall 2026/2025). Course material appearing on the trending page is itself a signal: agentic development practices have moved fast enough to become university curriculum within roughly the same window this series has been tracking them.

Open-Dev-Society/OpenStock (16,886 stars, +755 today): an open-source alternative to paid market-data platforms, tracking real-time prices and company insights for free. It sits in the same lineage as earlier “free, open alternative to an expensive incumbent” stories this series has flagged repeatedly. The pattern hasn’t slowed down, just changed which incumbent gets targeted.

The Rest of the Board

• affaan-m/ECC: 263,813 stars, +826 today. Agent harness performance optimization system
• anthropics/claude-code: 147,168 stars, +419 today. Anthropic’s own agentic coding tool
• addyosmani/agent-skills: 97,733 stars, +736 today. Production-grade agent skills
• paperless-ngx/paperless-ngx: 45,601 stars, +57 today. Document management system
• anthropics/financial-services: 35,403 stars, +260 today. Anthropic’s vertical entry into financial services
• trycua/cua: 25,190 stars, +1,018 today. Computer-use 2.0, drivers, fleets, benchmarks
• cloudflare/security-audit-skill: 18,110 stars, +2,428 today. Multi-phase security audit skill, fastest mover
• vercel-labs/json-render: 17,372 stars, +291 today. Generative UI framework
• Open-Dev-Society/OpenStock: 16,886 stars, +755 today. Open-source market-data platform alternative
• coder/coder: 16,082 stars, +379 today. Secure environments for developers and their agents
• higgsfield-ai/higgsfield: 5,406 stars, +465 today. GPU orchestration for trillion-parameter training
• BuilderIO/agent-native: 5,260 stars, +98 today. Framework for building agentic apps
• mihail911/modern-software-dev-assignments: 4,578 stars, +172 today. Stanford course assignments on modern software dev

What This Snapshot Actually Says

Four days without a snapshot turned out to be enough time for a distinct pattern to harden: agent infrastructure is no longer described case by case. It has its own shared vocabulary now (environments, harnesses, native frameworks, computer-use fleets) and its own recognizable shape on the trending page. Anthropic entering financial services as a vertical, rather than staying purely horizontal, and a security skill more than doubling in four days on the strength of verifiable output, both point the same direction: the ecosystem has stopped proving agents work and started building the specialized layers underneath and around them.

Get Baozilla, Let's go!’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

Baozilla Tech Daily

Key Statistics (September 21, 2026)

• affaan-m/ECC: 263,813 stars (+826/day). Still the largest repo tracked
• anthropics/claude-code: 147,168 stars (+419/day). Steady climb, slower base rate
• addyosmani/agent-skills: 97,733 stars (+736/day)
• paperless-ngx/paperless-ngx: 45,601 stars (+57/day)
• anthropics/financial-services: 35,403 stars (+260/day), new. Anthropic’s first vertical-specific repo in this series
• cloudflare/security-audit-skill: 18,110 stars (+2,428/day). More than doubled in four days, today’s fastest mover

Four-day deltas (Sept 17 to Sept 21):

• cloudflare/security-audit-skill: 7,380 to 18,110 (+10,730, ~2,680/day average)
• affaan-m/ECC: 260,330 to 263,813 (+3,483, ~870/day average)
• addyosmani/agent-skills: 95,477 to 97,733 (+2,256, ~564/day average)
• anthropics/claude-code: 145,540 to 147,168 (+1,628, ~407/day average)

Critical Signal: Agent infrastructure has consolidated into a recognizable category on its own terms.

• Secure agent environments (coder), computer-use fleets (cua), agent-native app frameworks (agent-native), and cross-harness optimization (ECC) are no longer described as generic tooling. They’re being named and trended as a distinct layer.
• Anthropic’s move into a financial-services vertical shows the same specialization instinct applied to industry, not just infrastructure.
• Cloudflare’s security skill doubling in four days shows how fast a vertical, verifiable agent skill can scale once trust in its output is established.
• Together these suggest the ecosystem’s center of gravity has shifted: from proving agents work, to specializing what they work on and what they run inside of.

Keywords for 20260921:

1. Agent-Infra
2. Financial-Services
3. Security-Audit
4. Steady-Growth
5. ECC