Devour/SKILL.md

name, description, allowed-tools

name	description	allowed-tools
devour	Use this skill for Devour CLI workflows: scrape docs, get language docs, query local index, ask docs-grounded questions, sync sources, run quality triage, and verify live smoke checks. Trigger on: "devour", "docs to ai", "scrape docs", "ask docs", "query docs", "sync docs", "quality scan".	Read Write Edit Glob Grep Bash WebFetch

Devour Skill

Use this skill when a task is explicitly about Devour CLI operations or troubleshooting Devour workflows.

What Devour now supports

• devour init
• devour get
• devour scrape
• devour scrape --sources ...
• devour query
• devour ask
• devour sync
• devour status
• devour push <path> (local ingest)
• devour serve (local stdio JSON-RPC)
• devour auto
• devour verify smoke
• devour quality ... (delegated to desloppify)
• devour review ... (delegated to desloppify review)

Remote server/push workflows are experimental.

Fast routing

1. User gives URL/source: use devour scrape.
2. User gives language+topic: use devour get.
3. User asks a question: use devour ask --lang ....
4. User wants local search: use devour query.
5. User wants updates from config: use devour sync.
6. User wants automatic intent routing: use devour auto.
7. User wants confidence check: use devour verify smoke.

Reliable workflow

devour init
devour get go net/http
devour query "http client timeout"
devour ask --lang go "how to parse json"
devour sync
devour status

Key behavior notes

• ask is hybrid local-first with targeted live fallback.
• query is local lexical index; no API key required.
• scrape fails by default when 0 docs are extracted (unless --allow-empty).
• serve local mode uses JSON-RPC over stdio.

Supported language aliases

• go, golang
• rust
• python, py
• java
• spring
• typescript, ts
• react
• vue
• nuxt
• docker
• cloudflare, cf
• astro
• csharp, cs
• kotlin, kt
• php
• ruby, rb
• elixir, ex
• next, nextjs
• svelte
• angular, ng
• remix
• solid, solidjs
• express, expressjs

Response expectations

When reporting command results:

1. Show exact command(s) run.
2. Summarize key output.
3. Show output file locations.
4. Call out limitations/experimental behavior.
5. Give the next command to continue.