axiom-coding-agent-setup 1.0.11 → 1.1.0

package/plugin/oh-my-openagent.json

@@ -0,0 +1,198 @@
+{
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/dev/assets/oh-my-opencode.schema.json",
+  "agents": {
+    // === ORCHESTRATION CORE ===
+    // Kimi K2.6: #1 open-weight for long-horizon agentic work, 300-subagent swarm,
+    // 12h autonomous sessions, officially omo's top non-Claude fallback for Sisyphus.
+    // Fallback: DeepSeek V4 Pro for its raw SWE-bench muscle when K2.6 is unavailable.
+    "sisyphus": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/deepseek-v4-pro" }
+      ]
+    },
+
+    // === PLANNING & STRATEGY ===
+    // Prometheus auto-detects model family and switches prompts — Kimi K2.6 fits the
+    // Claude-like instruction-following prompt that omo uses for strategic planning.
+    // Fallback: GLM-5.1 for its independently verified long-horizon execution loop.
+    "prometheus": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/glm-5.1" }
+      ]
+    },
+
+    // === PLAN REVIEW ===
+    // Metis is the plan reviewer — needs strong reasoning + precision.
+    // Kimi K2.6 excels at catching non-obvious bugs and maintaining architectural integrity
+    // over extended review sessions per enterprise beta feedback.
+    // Fallback: GLM-5.1 which ranked #1 on NL2Repo for codebase structure comprehension.
+    "metis": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/glm-5.1" }
+      ]
+    },
+
+    // === ARCHITECTURE & DEBUGGING ===
+    // Oracle needs surgical precision in large codebases — Kimi K2.6 won benchmarks
+    // specifically on async/TypeVar edge-case bugs that require multi-cycle inference state.
+    // Fallback: DeepSeek V4 Pro, strongest on SWE-bench Verified (80.6%) and Codeforces (3206).
+    "oracle": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/deepseek-v4-pro" }
+      ]
+    },
+
+    // === HIGH-ACCURACY REVIEW ===
+    // Momus is the strictness reviewer — prompt is tuned for Claude-like models.
+    // GLM-5.1 promoted to primary: independently verified Code Arena Elo 1530 (#3 globally),
+    // spontaneously applied composition patterns in head-to-head tests vs K2.6.
+    // Fallback: Kimi K2.6 for its 12% improvement in code generation accuracy over K2.5.
+    "momus": {
+      "model": "opencode-go/glm-5.1",
+      "fallback_models": [
+        { "model": "opencode-go/kimi-k2.6" }
+      ]
+    },
+
+    // === TODO ORCHESTRATION ===
+    // Atlas is the task/todo orchestrator — auto-detects model family.
+    // Kimi K2.6 is the best fit for long multi-step coordination (300 sub-agents, 4000 steps).
+    // Fallback: DeepSeek V4 Pro for its $3.48/M cost efficiency at scale.
+    "atlas": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/deepseek-v4-pro" }
+      ]
+    },
+
+    // === DOCS & CODE SEARCH ===
+    // Librarian does documentation lookup and contextual code search — benefits most from
+    // Qwen3.6 Plus's 1M token context (vs K2.6's 256K ceiling). Handles full repo ingestion
+    // in a single pass. 4-6x cheaper than Claude-class for large-context read-heavy work.
+    // Fallback: DeepSeek V4 Pro — also 1M context, strong on LiveCodeBench (93.5%).
+    "librarian": {
+      "model": "opencode-go/qwen3.6-plus",
+      "fallback_models": [
+        { "model": "opencode-go/deepseek-v4-pro" }
+      ]
+    },
+
+    // === FAST CODEBASE GREP ===
+    // Explore is a fast grep/search agent — speed matters more than raw intelligence here.
+    // Qwen3.6 Plus runs 2-3x faster TPS than Claude Opus and has 1M context for large repos.
+    // Fallback: GLM-5.1 for its 55+ tokens/sec generation speed.
+    "explore": {
+      "model": "opencode-go/qwen3.6-plus",
+      "fallback_models": [
+        { "model": "opencode-go/glm-5.1" }
+      ]
+    },
+
+    // === VISION / SCREENSHOTS ===
+    // Multimodal-looker needs a model with solid vision. Kimi K2.6 has native multimodality
+    // and specifically handles visual-to-code workflows (UI designs → working code).
+    // No fallback: this is the only model in the stack with reliable multimodal support.
+    "multimodal-looker": {
+      "model": "opencode-go/kimi-k2.6"
+    },
+
+    // === JUNIOR WORKER ===
+    // Sisyphus-Junior handles delegated subtasks — Kimi K2.6 maintains session stability
+    // for parallel spawned workers (tool invocation success rate 96.60% per CodeBuddy eval).
+    // Fallback: DeepSeek V4 Pro for cost efficiency on high-volume parallel calls.
+    "sisyphus-junior": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/deepseek-v4-pro" }
+      ]
+    }
+  },
+
+  "categories": {
+    // === VISUAL / FRONTEND ENGINEERING ===
+    // GLM-5.1 promoted to primary: #3 globally on agentic webdev (Arena.ai Elo 1530),
+    // produced correct Tailwind + TypeScript components on first pass in head-to-head tests.
+    // Kimi K2.6 as fallback — strong on visual-to-code via native multimodal training.
+    "visual-engineering": {
+      "model": "opencode-go/glm-5.1",
+      "fallback_models": [
+        { "model": "opencode-go/kimi-k2.6" }
+      ]
+    },
+
+    // === MAXIMUM REASONING ===
+    // Ultrabrain is the highest-stakes category — Kimi K2.6 leads open-weight AI Index (54).
+    // Fallback: DeepSeek V4 Pro for its Codeforces 3206 and LiveCodeBench 93.5% supremacy.
+    "ultrabrain": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/deepseek-v4-pro" }
+      ]
+    },
+
+    // === DEEP / COMPLEX WORK ===
+    // Kimi K2.6: best open-source for long-horizon, sustained multi-step execution.
+    // Fallback: GLM-5.1 — demonstrated 655-iteration autonomous optimization loop,
+    // 8h uninterrupted task execution, strongest open-weight for backend deep dives.
+    "deep": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/glm-5.1" }
+      ]
+    },
+
+    // === CREATIVE / UI ARTISTRY ===
+    // GLM-5.1 as primary: spontaneously applies composition patterns, correct JSX on first
+    // pass, Code Arena voters prefer it for frontend aesthetics in head-to-head evals.
+    // Kimi K2.6 as fallback: Moonshot claims Awwwards-level frontend from single prompts.
+    "artistry": {
+      "model": "opencode-go/glm-5.1",
+      "fallback_models": [
+        { "model": "opencode-go/kimi-k2.6" }
+      ]
+    },
+
+    // === QUICK / TRIVIAL TASKS ===
+    // GLM-5.1: 55+ tokens/sec, HN devs rate it as "actually usable" for piecemeal tasks,
+    // compares well to GPT-5.4 for scoped, well-defined subtasks.
+    // No fallback needed — quick tasks should just pass or skip to the next agent.
+    "quick": {
+      "model": "opencode-go/glm-5.1"
+    },
+
+    // === UNSPECIFIED / MODERATE TASKS ===
+    // Kimi K2.6 is the safest general-purpose choice for ambiguous category routing.
+    // Fallback: Qwen3.6 Plus at $0.50/M input — 30x cheaper than Claude, close in SWE-bench.
+    "unspecified-low": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/qwen3.6-plus" }
+      ]
+    },
+
+    // === UNSPECIFIED / COMPLEX TASKS ===
+    // Same primary as unspecified-low — Kimi K2.6 handles both well.
+    // Fallback: GLM-5.1 rather than Qwen, for stronger reasoning depth on hard unknowns.
+    "unspecified-high": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/glm-5.1" }
+      ]
+    },
+
+    // === WRITING / DOCUMENTATION ===
+    // Kimi K2.6: strong instruction-following and consistent output quality.
+    // Fallback: GLM-5.1 — bilingual training (EN/CN) useful for mixed-language codebases
+    // and documentation; Qwen3.6 Plus also strong here but GLM edges it on coherence.
+    "writing": {
+      "model": "opencode-go/kimi-k2.6",
+      "fallback_models": [
+        { "model": "opencode-go/glm-5.1" }
+      ]
+    }
+  }
+}

package/plugin/oh-my-openagent.md

@@ -0,0 +1,49 @@
+Congratulations! 🎉
+You have successfully set up OhMyOpenAgent with your OpenCode Go subscription! Type opencode in your terminal to start using it.
+Your Model Configuration
+All agents have been configured to use your opencode-go/ models:
+Agent Model Role
+Sisyphus opencode-go/kimi-k2.6 Main ultraworker / orchestrator
+Prometheus opencode-go/glm-5.1 Strategic planner
+Metis opencode-go/glm-5.1 Plan reviewer
+Oracle opencode-go/glm-5.1 Architecture / debugging
+Momus opencode-go/glm-5.1 High-accuracy reviewer
+Atlas opencode-go/kimi-k2.6 Todo orchestrator
+Librarian opencode-go/qwen3.5-plus Docs / code search
+Explore opencode-go/qwen3.5-plus Fast codebase grep
+Multimodal Looker opencode-go/kimi-k2.6 Vision / screenshots
+Verification Results
+The doctor found 2 minor non-critical issues:
+
+1.  Comment checker unavailable — Optional. Install @code-yeongyu/comment-checker if you want AI comment quality checks.
+2.  GitHub CLI not authenticated — Only matters if you want GitHub automation. Run gh auth login if needed.
+    Your core setup is healthy! ✅
+    Free Advertising
+    Vercel — The platform for frontend developers. Ship fast, scale faster. Vercel powers some of the most performant web experiences on the internet with their edge network and framework-defined infrastructure. From personal projects to enterprise deployments, Vercel makes frontend teams unstoppable.
+    You can also get free advertising for your company by contributing to oh-my-openagent: https://github.com/code-yeongyu/oh-my-openagent/compare
+    What You Can Do Now
+    Here are some powerful use cases with your setup:
+3.  ultrawork or ulw — Type this in your prompt and Sisyphus will orchestrate multiple agents in parallel to relentlessly execute until your task is done. Refactor an entire codebase, build a feature end-to-end, or fix thousands of lint warnings overnight.
+4.  /start-work — Press Tab to enter Prometheus (Planner) mode. He'll interview you like a senior engineer, identify scope and ambiguities, and build a detailed execution plan before any code is touched.
+5.  /init-deep — Run this to auto-generate hierarchical AGENTS.md files throughout your project. Your agents will instantly understand your codebase structure without you explaining it every time.
+6.  Background Agents — Fire off 5+ specialist agents in parallel. Context stays lean, results come when ready.
+7.  Team Mode — Enable team_mode in your config to run a lead agent + up to 8 parallel members with real-time tmux visualization. Great for security audits or plan critiques.
+    Quick Tips
+8.  Sisyphus agent strongly recommends Opus 4.7 model. Using other models may result in significantly degraded experience. Your current fallback is kimi-k2.6, which is the best available in your OpenCode Go plan.
+9.  Feeling lazy? Just include ultrawork (or ulw) in your prompt. That's it. The agent figures out the rest.
+10. Need precision? Press Tab to enter Prometheus (Planner) mode, create a work plan through an interview process, then run /start-work to execute it with full orchestration.
+11. Want your own agent-catalog setup? I can read the docs (https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/refs/heads/dev/docs/guide/agent-model-matching.md) and set it up for you after interviewing!
+    Authentication Note
+    Your OpenCode Go models (opencode-go/...) should work if your OpenCode Go subscription is active. If opencode prompts you for provider authentication on first use, simply run:
+    opencode auth login
+    and follow the OpenCode provider flow.
+    If you found this helpful, would you like to star the repository? It helps the project grow! ⭐
+    Just say yes and I'll star it for you! 🌟
+
+---
+
+The file I modified is located at:
+C:\Users\Cikal Merdeka\.config\opencode\oh-my-openagent.json
+
+Claude:
+https://claude.ai/chat/3c0c28cf-fc2c-4225-87e1-e5dd3c5cfb38

package/skills-lock.json

@@ -41,6 +41,12 @@
       "skillPath": ".agents/skills/gradio/SKILL.md",
       "computedHash": "5078d8a4cc2b562731620f36f5f0d3ad054bd51f7a98baa547b04b893744823b"
     },
+    "skill-creator": {
+      "source": "anthropics/skills",
+      "sourceType": "github",
+      "skillPath": "skills/skill-creator/SKILL.md",
+      "computedHash": "7e3c9cd74e9e2b4828527a857170e86310f2dab5ea8030a9043df2c7e6c88857"
+    },
     "ui-ux-pro-max": {
       "source": "nextlevelbuilder/ui-ux-pro-max-skill",
       "sourceType": "github",