auto-researchlisted

# Auto-Research Autonomous goal-driven orchestration inspired by Karpathy's autoresearch pattern. The user writes intent in plain English; the agent classifies the goal, selects the right skill, configures parameters, and runs the appropriate autonomous loop. **Core principle:** Human writes intent, agent figures out execution. One entry point, many execution paths. ## When to Use | User says... | Auto-research routes to | |---|---| | "fix the flaky test in auth" | `/fix-bug` | | "improve API latency below 100ms" | `/experiment --minimize` | | "get coverage to 90%" | `/cover --target 90` | | "design a caching layer" | `/brainstorming` | | "build the notification feature" | `/develop` | | "review MR !42" | `/review-mr` | | "make sure everything passes" | `/verify` | | "optimize the review skill prompt" | `/experiment` on SKILL.md | | "what's the state of MCP auth in 2026" | `/ctk:web-research` | > **Illustrative examples only.** The authoritative routing map — categories, signal words, and target skills — is the **Intent Classification table** in Phase 1 (§Classify). Change routing *there*; this table just shows it in practice. **Use `/auto-research` instead of a specific skill when:** - The user describes a goal, not a method - The right skill is not obvious from the request - The user wants the agent to pick the approach **Use a specific skill directly when:** - The user already knows which skill to use - The request maps unambiguously to one skill ### Routing Decisi