Amazon Q vs Claude Code
AQ
- similar to CLAUDE.md, uses amazonqrules
- indexes workspace codebase as a whole
- workspace index and amazonqrules are complementary in which rules are just for consistency in the queries
CC
- does not index the whole codebase but searches it on the fly when querying
- uses CLAUDE.md for context
Claude Skills - https://www.youtube.com/watch?v=421T2iWTQio&t=718s
- problem is running out of context window
- claude skills solve this by not loading all of the context
- skills are just folder with SKILL.md file
- plus other context files
- Progressive disclosure - token efficient since skills only inpyt Name+Description. The agent then decides if it want to use the skills.
- Slash Commands (triggered by user) vs skills used automatically
- MCPS uses the full context load, infrastructure provisioning
Personal take on skills: I have not reached a problem wherein I needed the full context window. I only use LLM for small tasks such as creating functions.