Semantic codebase retrieval using the Augment Context Engine through a small ace command.
Use it when the relevant identifiers are unknown and the query is about behavior, intent, architecture, or data flow. Treat results as candidate files, then read the returned files and confirm exact identifiers or call sites with literal search.
Before searching, you must authenticate using one of the following methods:
If you have an official Augment account:
auggie loginIf you use an ace relay middleman service, you can skip auggie login by manually configuring ~/.augment/session.json with relay-specific settings:
accessToken: Use the relay token (e.g., starts withace_).tenantURL: Change to the relay endpoint (e.g.,https://acemcp.heroman.wtf/relay/).scopes: Typically set to["email"].
From this repository:
uv tool install .From a published package:
uv tool install ace-wrappertimeout 60s ace "How is the authentication flow implemented?" -w /path/to/project/root
timeout 60s ace "Find where uploaded files are validated and stored" -w /path/to/project/root
timeout 60s ace "用户登录流程在哪里实现" -w /path/to/project/rootUse --verbose for debug logs:
timeout 60s ace "config loading" -w /path/to/project/root --verboseTo use ace effectively with AI coding agents, refer to references/prompts.md for recommended prompt instructions and best practices.
- ACE is a candidate-file generator, not proof that a feature exists.
- Verify important conclusions by reading files and using exact search.
- Split unrelated workflows into separate queries.
- Use literal search for exact identifiers, error strings, routes, events, and config keys.
Detailed usage guidance and AI Agent prompt templates live in references/.
Publishing is handled by GitHub Actions. Create and push a version tag to build, publish to PyPI, and create a GitHub Release:
git tag v0.1.1
git push origin v0.1.1The PyPI publish job uses Trusted Publishing, so the PyPI project must allow the ferstar/ace-wrapper repository and the pypi GitHub environment.