Trove / Get started
Installation
Build the trove engine from source with Cargo and put it on your PATH, and install the Trove plugin in Claude Code from its marketplace or by hand.
On this page
The engine and the plugin install separately. You can use either one on its own; they work best together.
First principles
The engine is a Rust program and there are no prebuilt binaries, so installing it means compiling it. The plugin is a folder of instructions and Bash scripts that Claude Code reads, so installing it means pointing Claude Code at that folder. The two meet in one place: the plugin’s index and ship skills call the trove binary and look for it on your PATH first. Putting trove on your PATH is therefore the step that makes the engine work in any repository, not only inside the Trove clone.
Engine
Requirements
- A Rust toolchain with Cargo. Continuous integration (CI) builds with stable Rust on Ubuntu and macOS. The workspace uses the 2021 edition.
- A C compiler. The tree-sitter grammars are compiled from C sources during the build.
- Git, to clone the repository.
- Optional:
curlon yourPATH, only fortrove ask --provider anthropicorollama. The binary sends Hypertext Transfer Protocol (HTTP) requests through the systemcurlinstead of linking its own Transport Layer Security (TLS) stack.
Build
git clone https://github.com/anishfyi/trove
cd trove
cargo build -p trove-cli
This builds the trove binary (package trove-cli) into target/debug/trove. Use cargo build --release -p trove-cli for an optimized build in target/release/trove. cargo build --workspace also builds trove-ship, a second binary with the same flags as trove ship.
Check the build:
cargo run -p trove-cli -- --version
This prints trove 0.2.0. Both the v0.2.0 release and the current main branch report that version.
Run it from the clone or put it on your PATH
From inside the clone, every command can run through Cargo:
cargo run -p trove-cli -- status
cargo run only works inside the Trove workspace, and by default the engine works on the current directory, which is then the Trove clone itself. To work on another repository, pass --repo, or call the built binary directly:
./target/debug/trove --repo /path/to/your/repo index
The better option is to install the binary into ~/.cargo/bin, so trove works from any directory and the plugin skills can find it:
cargo install --path crates/trove-cli
Update
From the clone:
git pull && cargo install --path crates/trove-cli
Then re-run trove index in each repository you index. Re-indexing rewrites the summaries in .trove/.
Plugin
Requirements
- Claude Code with plugin support.
- Bash, Git and the usual core utilities, for the hook and helper scripts.
- Optional:
jqorpython3, fortrove.sh context, which turns the engine index into the project trove’s context section.
Install from the marketplace
Run these one at a time inside Claude Code.
-
Add the marketplace. A marketplace is a Git repository that lists installable plugins. This points Claude Code at the Trove repository. Nothing is installed yet.
/plugin marketplace add anishfyi/troveIf you see a Secure Shell (SSH)
Permission denied (publickey)error, you have no SSH key on GitHub. Use the HTTP Secure (HTTPS) address instead:/plugin marketplace add https://github.com/anishfyi/trove.git -
Install the plugin. Read the argument as
plugin@marketplace:troveis the plugin andvelofy-troveis the marketplace name declared in.claude-plugin/marketplace.json. This installs seven skills (init,remember,recall,index,query,render,ship) and the SessionStart and SessionEnd hooks./plugin install trove@velofy-trove -
Reload so it activates. A newly installed plugin is not live until you reload. Confirm it with
/plugin list, and type/trove:to see the commands. The SessionStart hook runs from your next new session onward./reload-plugins -
Create your trove. See Plugin quickstart.
/trove:init
The marketplace is named velofy-trove even though the repository lives at github.com/anishfyi/trove. Up to v0.2.0 it was named anishfyi-trove; it was renamed on 2026-09-12. If you added the marketplace before then, it may still be listed under the old name.
Manual install (skills only, no hooks)
git clone https://github.com/anishfyi/trove /tmp/trove
cp -r /tmp/trove/plugins/trove/skills/* ~/.claude/skills/
This copies the seven skill folders into ~/.claude/skills/. Skills installed this way are not namespaced by the plugin, so they appear as /init, /remember, /recall and so on. Two things only come with the plugin install:
- The SessionStart and SessionEnd hooks.
${CLAUDE_PLUGIN_ROOT}. The skills call the helper as"${CLAUDE_PLUGIN_ROOT}/scripts/trove.sh", and Claude Code sets that variable only for plugins. With a manual copy, tell Claude to runplugins/trove/scripts/trove.shfrom your clone instead.
Update
Run /reload-plugins after the marketplace copy updates.
Using both together
The index and ship skills look for the engine in this order:
- The
trovebinary on yourPATH. - Inside a checkout of the Trove repository,
cargo run -p trove-cli --. - In any other repository, Cargo pointed at a Trove checkout:
cargo run --manifest-path <trove>/Cargo.toml -p trove-cli -- --repo ..
The query skill’s instructions still call cargo run -p trove-cli directly, which only resolves inside a Trove checkout. With trove on your PATH, run trove query yourself or ask Claude to use the binary. See Plugin reference.