What problem does it solve? Coding agents sometimes miss project conventions because AGENTS.md context fails to surface, or they pick the wrong tool — reading files via shell cat, using grep where ripgrep is faster — causing timeouts and missed instructions. This Skill documents how context injection works and which tool to use when. ## Core Features & Use Cases - AGENTS.md injection mechanics: Explains the primary path (read AGENTS.md after setting cwd) and the fallback path (any file read scans upward to git root or $HOME), including dedup state and auto-reset after compaction. - Shell vs files tool choice: Clarifies that git, grep, tests, and builds go through ava.shell.run while known single-file reads and writes use ava.files, with common mistakes called out. - ripgrep over grep: Shows benchmark data proving rg is roughly 50x faster than grep -rn on repos with many .worktrees checkouts, plus fallback grep flags when rg is unavailable. - Use Case: Your agent just ran cat src/main.py via shell and never saw the project's AGENTS.md rules. This Skill explains why shell reads bypass injection and tells you to use ava.files.read instead. ## Quick Start Ask the agent why AGENTS.md conventions are not appearing and whether to use shell or file tools for reading source code.
