Reusable preferences for your AI agent.
Traits turns feedback into preferences your agent can use across tasks and
sessions. Each trait is a Markdown file in .agents/traits/<name>.md:
---
name: direct-callbacks
subject: Callbacks passed to functions.
requirement: >
Pass compatible functions directly instead of wrapping them in forwarding
arrows.
purpose: Expose callback behavior without redundant wrappers.
---Your agent proposes traits, asks you to approve changes, and checks its work against them through MCP. The files live in your project, so you can edit, review, and share them with your team.
Paste this into your agent inside your project:
Set up Traits using https://github.com/crosshatch/traits/blob/main/SETUP.md.
Follow this project's conventions, ask me to choose providers and models, and
preserve existing configuration.
check_traits returns { traits, files }, grouped by file for coordinated
repairs. Each referenced trait definition appears once. Each file includes its
path, full-snapshot SHA-256 revision, and source chunks; each chunk contains its
text once and findings that reference trait names. Classifications and
probabilities are preserved. Chunk start lines are not exact violation
locations.
No findings returns { "traits": [], "files": [] } for the checked selection.