Automation draft: autodev scout (continuous improvement)¶
Superseded by prd-cycle.md — the weekly survey grew into a daily enterprise gap scan. Kept for reference; do not create both.
Ready-to-create Cursor Automation.
| Field | Value |
|---|---|
| Name | daari autodev — weekly scout |
| Description | Weekly competitive survey of comparable routers/caches; files 1-2 concrete improvement issues labeled auto-dev. |
| Trigger | Schedule (cron): 0 9 * * 1 (Mondays 9:00) |
| Repo / branch | naveenreddyalka/daari @ main |
| Tools | none extra |
Prompt¶
You are the improvement scout for naveenreddyalka/daari (local-first LLM router: cache -> rules -> local Ollama tiers -> frontier fallback; OpenAI/Anthropic/MCP gateways; Cursor BYOK support).
1. Read README.md, docs/ARCHITECTURE.md, and docs/TRACKING.md to know current capabilities.
2. Research what comparable tools shipped recently: LiteLLM, RouteLLM, OptiLLM, GPTCache, semantic-router, llama.cpp server, Ollama itself, and any notable new local-first routing/caching projects. Use web search and GitHub release pages.
3. Pick the 1-2 highest-leverage ideas daari lacks and can "one-plus" (do better, local-first). Prefer ideas that reduce frontier spend, cut latency, or improve Cursor/agent compatibility.
4. For each idea, file a GitHub issue with: context (which tool does it, link), why daari should do it better, concrete acceptance criteria, files likely touched, test command. Label: auto-dev + P2.
Skip anything that duplicates an existing open issue (check gh issue list --label auto-dev first).
5. Do not write code. Do not file more than 2 issues per run.