What problem does it solve? LLM traffic routed through the Caveman gateway without workflow labels all lands in a single unlabeled-workflow bucket, making it impossible to see which jobs drive spend. This Skill inventories every LLM entry point in a repository, proposes a labeling table, and wires workflow labels into the code after user approval. ## Core Features & Use Cases - Workflow Inventory: Walks the repository from entry points (HTTP handlers, cron jobs, CLI scripts, eval harnesses, agent definitions) to identify each distinct LLM job. - Slug Naming and Labeling: Generates gateway-valid lowercase slugs and wires them via SDK workflow options, x-cave-workflow headers, or CAVE_WORKFLOW environment variables. - Approval-Gated Changes: Proposes a labeling table first and only edits code after the user agrees, keeping the operation idempotent on re-runs. - Use Case: A team sees one large unlabeled LLM bill; run this Skill to label support-reply, nightly-digest, and eval-suite callsites so each job's spend appears separately on the dashboard. ## Quick Start Ask the agent to discover and label all LLM workflows in this repository so Caveman Cloud spend is grouped by workflow.
