AI Playbooks

VETA accepts contributions from human engineers and from LLM agents (Claude on a cloud session, qwen3-coder on the Spark workstation). The Astro docs are the golden source for how the platform works; these playbooks are the golden source for how to extend it.

Each playbook captures a recurring task — adding an algo, debugging a Grafana panel, fixing an accessibility violation — with the load-bearing context an agent needs to land a correct, well-tested change without re-discovering it every time.

When to write a new playbook

Three rules of thumb:

• The task has been done at least twice with similar friction.
• An agent without prior session context would benefit from the recipe.
• The recipe is shorter than re-reading 300 lines of source code.

Available playbooks

Add a new algo strategy Where the strategy lives, what events it must produce/consume, how to make it pass tests + show up in docs.

Add a new dashboard panel panelRegistry + layoutModels + tests. The minimum surface to make a panel render and ship.

Debug an empty Grafana panel Diagnostic order-of-operations for 'No data' in a provisioned dashboard panel.

Playbook shape

Each playbook follows a consistent structure so an agent (or a human in a hurry) can skim:

1. When to use — concrete trigger conditions
2. Inputs needed — files, env vars, services that must be reachable
3. Steps — numbered, with file paths and code shapes
4. Verification — exactly which test invocations prove success
5. Common pitfalls — what reliably breaks first time

This is also the shape used by the runtime playbooks the LLM-advisory service loads from backend/src/llm-advisory/playbooks/. Different surface (TypeScript module vs Markdown page) but the same intent: capture working knowledge once, reuse forever.