ECC Surges on GitHub as Cross-Harness Agent OS
ECC, the open-source agent harness operating system maintained by GitHub user affaan-m, surged onto early September 2026 trending lists with roughly 254,000 stars and daily gains reported above
PromptCrates Editorial
Staff Writer
ECC, the open-source agent harness operating system maintained by GitHub user affaan-m, surged onto early September 2026 trending lists with roughly 254,000 stars and daily gains reported above 400 stars on public trackers, cementing its status as one of the most watched cross-harness toolkits for Claude Code, Codex, Cursor, OpenCode, and related coding agents. The MIT-licensed project packages skills, specialized agents, commands, hooks, memory, MCP inventory, and worktree lifecycle controls so teams can share one engineering system across multiple agent runtimes instead of rewriting prompts per tool. For developer-tooling watchers, ECC is a news story about infrastructure consolidation in the agent era—not a tutorial—because its star velocity and multi-harness adapters signal where open-source agent ops is concentrating.
Why ECC is trending now
GitHub's public page for affaan-m/ECC lists on the order of 254,000 stars, and community trending badges have highlighted hundreds of new stars per day—often cited around 411 stars per day during the early September spike. That scale puts ECC in rare company among agent-adjacent repositories and explains why news desks are treating the repo as a market signal rather than a niche plugin.
Positioning matters. ECC markets itself as an agent harness performance and coordination layer: planning before build, test verification, fresh-context review, memory of what worked, and conversion of repeated wins into reusable skills. Version 2.0.0 arrived in June 2026, with 2.2 guidance in August expanding packaged setup for Claude Code, Codex, and Kimi Code. The September trending wave therefore looks less like a first launch and more like network effects after the 2.x surface stabilized across harnesses.
The project's own inventory tables describe on the order of 64 specialized agents for planning, review, build repair, security, and architecture, plus roughly 261 skills spanning TDD, research, security, docs, frontend, data, ML, and operations, alongside about 84 legacy command shims while the surface shifts skills-first. Hooks, session memory, continuous learning, and AgentShield scanning for prompts, MCP config, permissions, and secrets round out the runtime story. Exact counts move with weekly releases; the durable news point is the breadth of a single MIT repo trying to be the shared OS under several commercial agent CLIs.
Readers tracking neighboring open agent tooling can compare ECC's surge with our earlier notes on OpenAI Codex CLI GitHub trending and Ponytail GitHub trending agent skills, two other September signals that skills and CLI harnesses—not only chat UIs—are driving developer attention.
Cross-harness adapters and product split
ECC's strategic bet is portability. Claude Code remains the deepest native path via the ecc@ecc plugin and guided npx ecc-universal setup, while Codex gains a native marketplace plugin flow and a deprecated legacy sync path. Capability-limited adapters cover Cursor, OpenCode, Gemini, Zed, GitHub Copilot, Antigravity, Qwen, and other environments, with an explicit support-status matrix warning that feature parity is incomplete. That honesty is part of the news: ECC is not claiming one binary runs identically everywhere; it is claiming a shared skill and agent library with harness-specific install edges.
Commercially, the open-source core stays MIT forever while ECC Pro and a GitHub App monetize private-repo hosting, with public pricing references starting around $19 per seat per month for private work. Sponsors and Pro subscribers fund weekly shipping across multiple harnesses under a small maintainer footprint—a sustainability model other agent repos are watching as star counts race ahead of staffing.
For platform teams already juggling Claude Code, Codex, and Cursor licenses, the product split reframes procurement. Instead of buying three incompatible prompt packs, engineering leads can evaluate whether one MIT skill inventory plus paid private hosting reduces drift between agents that share the same codebase. That evaluation is especially sharp for companies that adopted Codex CLI during its own GitHub trending cycle and now need a skills layer that does not lock them to a single vendor shell.
Security buyers will also notice AgentShield and MCP inventory language sitting next to the star chart. As agent tooling grows more powerful, scanners that flag risky prompts, over-broad permissions, and secret leakage become part of the same news package as raw star velocity—echoing themes in our coverage of CrowdStrike SafeMind and NVIDIA Nemotron, where runtime guardrails for agentic systems moved from research talk into product announcements.
Inventory docs still put skills first—on the order of 261 skills, 64 agents, and about 84 command shims in mid-2026 snapshots—while hooks, worktree lifecycle controls, and MCP inventory sit beside AgentShield as the operational layer that turns a star chart into something platform teams can evaluate.
What developer platforms should watch next
Three follow-throughs will decide whether ECC stays a trending curiosity or becomes default infrastructure. First, whether Cursor, OpenCode, and Copilot adapters close enough of the parity gap that teams stop maintaining forked skill trees per harness. Second, whether Pro and App revenue keep pace with the cost of supporting weekly releases across that matrix under a small maintainer set. Third, whether AgentShield and MCP inventory become procurement checkboxes that enterprises cite when comparing ECC to thinner skill packs such as Ponytail.
Geographically, the star surge is global—GitHub trending lists do not stop at Silicon Valley—but the deepest native integrations still cluster around US-origin agent CLIs from Anthropic, OpenAI, and Cursor. European and Asian platform teams adopting ECC will need to decide how MIT skills interact with local data-residency rules when agents call cloud MCP servers. That is a governance question, not a setup walkthrough, and it belongs in the same board packet as license audits.
For 9 September readers, ECC is the clearest open-source signal yet that agent harnesses are consolidating around shared skills, hooks, and MCP inventories instead of one-off system prompts. The repository and its v2.0.0 tag are the primary sources; star velocity is the distribution story; cross-harness adapters are the strategic bet. Whether that bet holds depends on parity, funding, and security packaging—not on another tutorial thread.
