Curated Open-Source AI Repositories

“I maintain 41 skills under .claude/skills and .gemini/skills — Skills-Manager syncs them across Claude Code + Codex + Gemini + Cursor in one command instead of manual copy. Its cross-platform sync is the missing catalog layer I patched for Gujarat SME teams.”

“I reverse 26% vulnerable MCP skills per audit — wagner 14-skill RE pipeline standardizes that RE flow in Claude Code marketplace format. Its JS plugin is the RE template I graft onto my SkillSpector gate.”

“I ship QA swarms that cut bugs 87% in CI/CD — ios-agentic-skills brings that swarm to iOS/watchOS audit with consistent playbooks. Its QA harness is the iOS analog of my autonomous QA swarm pattern.”

“I gate Razorpay ₹15K+ via HITL where money moves — CeAlthubaiti agents apply the same OPA+HITL gate to Polymarket trades. Its strategy + opportunity loop is the market analog of my RFQ → OTel audit.”

“I run Claude Code + Codex + OpenCode harnesses where cross-harness delegation is manual — codex-plugin-cc lets Claude Code delegate review tasks to Codex natively. Its harness-bridge pattern is the interop layer my Junagadh multi-harness setup needed.”

“I test Google ADK vs LangGraph vs CrewAI for Gujarat subagents — munder-difflin is the local-first multi-agent harness that runs the same supervisor pattern without cloud lock-in. Its TS harness is the lightweight alternative I benchmark against ADK from Junagadh.”

“I gate every tool via OPA + HITL at the runtime layer — substrate builds that substrate as a Go core system instead of Python glue. Its systems-level agent runtime is the foundation pattern I study for Pi 5 edge agents at 62 tok/s.”

“I ship browser-use automation where agents stall on Cloudflare and logins — Cloudflare giving agents a first-class computer primitive removes the whole stealth-browser arms race. Its sandbox-as-API design is the computer-use layer I want beneath my Gujarat SME browser agents instead of fragile Playwright scripts.”

“I installed this in my Junagadh lab after agents kept dumping 400-line answers on small fixes. Responses got scannable in one day. It pairs well with my P95 ledger reviews where I skim 20 agent runs fast.”

“I ran the paper-trading loop for a Surat client who wanted research summaries, not auto-orders. The debate logs read like an analyst desk. I cap it at end-of-day reports on the VPS, never live execution.”

“I use this as a starter shelf for new Claude Code seats, then trim hard. The Postgres and GitHub MCP entries saved me an hour per setup. My rule: read before install, keep the analytics dashboard on.”

“I keep agent memory in git in my Junagadh lab, so git log on memory beats any dashboard. Branching a personality to test a hunch feels natural. I run it in-process with file and Postgres MCP servers attached.”

“I keep agent memory in git in my Junagadh lab so git log beats any dashboard. Ouroboros carrying identity plus durable memory across restarts is the exact continuity pattern I prototype for Curro, and I run it lab-only with file plus Postgres MCP servers attached for evals.”

“I wire Codex seats for Gujarat SME pilots from Junagadh, and reference plugins beat hand-rolled prompts for repeatable deploys. I copied the Notion capture pattern for a Rajkot client site docs setup and had capture running the same evening on the Rs 6K VPS. I keep the Figma plugin disabled by default — design tokens without a reviewer in the loop produced three off-brand landing drafts before I gated it.”

“I track token burn per run in my P95 ledger on fixed-price builds, so a 20-50 percent read saving is margin. I trialed Vix plan mode on a Laravel RFQ module in Junagadh and the stem-agent cache reuse across explore plus plan phases cut one review loop from $3.65 to $1.86. The 3,000-line-file case in their own benchmark still lags — I keep big-file refactors on my standard harness.”

“I maintain a 41-skill Junagadh catalog, and the official spec is what I measure third-party skills against. The document-skills section standardized how my Curro pipeline handles DOCX briefs for two Surat clients. I install with tight approvals after auditing trending skills for broad permission requests — the same gate I run before any skill touches payment keys.”