# skillnet
`skillnet` is a CLI for managing canonical AI skill stores, materialising derived agent views, and recording calibration data for `multi-phase-plan`.
The supported interface in `0.6.0` is the `skillnet` binary. This crate does not commit to a stable embeddable Rust API yet.
## Install
```sh
cargo install skillnet
```
For a source checkout:
```sh
cargo install --path .
```
## Hook ingestion
`skillnet hook install` wires Claude Code to record skill invocations through
`skillnet hook ingest`, storing rows in `skill_invocations`. Use
`skillnet hook status` to confirm the managed hook entries are installed. See
the mdBook chapter at [docs/src/hook-ingestion.md](docs/src/hook-ingestion.md)
for Postgres, SQLite, and Home Manager setup.
### Nix Home Manager
Add the flake input and import the module:
```nix
inputs.skillnet.url = "git+ssh://git@codeberg.org/caniko/skillnet.git";
# In your Home Manager config:
imports = [ inputs.skillnet.hmModules.default ];
programs.skillnet = {
enable = true;
settings = {
global = {
views = [
{ label = "claude"; path = "/home/alice/.claude/skills"; scope = "global"; }
{ label = "agents"; path = "/home/alice/.agents/skills"; scope = "global"; }
];
};
};
mirrorRoot = "/home/alice/skills-mirror";
database = {
backend = "postgres";
urlFile = config.age.secrets.skillnet-pg-url.path;
};
};
```
The module is exported as both `hmModules.default` and `hmModules.skillnet`.
It installs `skillnet` on `PATH`. When `settings` is declared, the module
renders `$XDG_CONFIG_HOME/skillnet/skillnet.toml`, which the CLI discovers by
default. When `catalogSettings` is declared, it renders
`$XDG_CONFIG_HOME/skillnet/skillnet.catalog.toml`. `skillsRoot`, `mirrorRoot`, and the
declarative database options are folded into generated `skillnet.toml`, so a
normal declarative Home Manager install does not depend on shell-specific env
imports. Without `settings`, you can still drop your own TOMLs and point
`programs.skillnet.configFile` at them. When SQLite is selected, the module also
creates the runtime data directory and exports `skillnet_DATA_DIR` and
`SKILLNET_DATA_DIR` for compatibility.
If you also want the module to define where the `ai-skills` checkout lives,
set `skillsRoot`. On atlas, that path is:
```nix
programs.skillnet.skillsRoot = "/data/nvme0/can/Projects/ai-skills";
```
When configured, the module writes `skills_root` into `skillnet.toml`, exports
`AI_SKILLS_REPO` for compatibility, and warns if the checkout directory is
missing. `skillsRoot` points at the skills checkout and VCS working tree;
`dataDir` remains skillnet's runtime database location.
Postgres is the default calibration backend and requires a connection URL.
SQLite is also supported by selecting it explicitly:
```nix
programs.skillnet = {
enable = true;
package = inputs.skillnet.packages.${pkgs.system}.skillnet;
database = {
backend = "sqlite";
};
};
```
The Postgres backend is included in default builds. If you override
`programs.skillnet.package` with a custom `--no-default-features` build, choose
a package that includes the `postgres` feature when using the Postgres backend.
The `database.url` value is written into the Nix store. For production secrets,
prefer `database.urlFile`, which reads the Postgres URL from a file at shell
initialization time.
Claude Code hook ingestion can also be installed declaratively:
```nix
programs.skillnet = {
enable = true;
hooks = {
enable = true;
events = [ "PostToolUse" ];
matchers = [ "Skill" ];
settingsFile = "${config.home.homeDirectory}/.claude/settings.json";
};
};
```
When enabled, Home Manager runs `skillnet hook install` during activation after
`writeBoundary`, so the change is idempotent and respects `home-manager switch
--dry-run`. Setting `hooks.enable = false` later leaves the settings file
untouched; run `skillnet hook uninstall` explicitly to remove managed entries.
Options:
- `programs.skillnet.dataDir` defaults to `${config.xdg.dataHome}/skillnet`.
This is skillnet's runtime database location.
- `programs.skillnet.skillsRoot` sets `skills_root` in generated
`skillnet.toml`, exports `AI_SKILLS_REPO` for compatibility, and is the
canonical mirror destination/VCS working tree; atlas uses
`/data/nvme0/can/Projects/ai-skills`.
- `programs.skillnet.mirrorRoot` is the legacy mirror destination option. When
both `skillsRoot` and `mirrorRoot` are set, they must match.
- `programs.skillnet.database.backend` selects `sqlite` or `postgres` and
defaults to `postgres`.
- `programs.skillnet.database.path` optionally sets the SQLite database path;
when unset, calibration data lives at
`<dataDir>/multi-phase-plan/calibration.sqlite`.
- `programs.skillnet.database.url` is written into generated `skillnet.toml`
when `settings` is declared, or exported as `SKILLNET_DATABASE_URL` for
user-managed config.
- `programs.skillnet.database.urlFile` reads a Postgres URL from a file at
shell initialization time.
- `programs.skillnet.settings` and `programs.skillnet.catalogSettings` render
`skillnet.toml` and `skillnet.catalog.toml` from Nix.
- `programs.skillnet.configFile` and `programs.skillnet.catalogConfigFile`
point the CLI at user-managed TOML files.
- `programs.skillnet.activation.promote`, `failOnConflict`, and `allowDelete`
control Home Manager's activation-time `skillnet sync` invocation. Promotion
is off by default for consumer-host safety.
- `programs.skillnet.hooks.enable` installs skillnet-managed Claude Code hook
entries in `programs.skillnet.hooks.settingsFile`.
- `programs.skillnet.hooks.events` and `programs.skillnet.hooks.matchers`
control the event/matcher pairs passed to `skillnet hook install`.
- `programs.skillnet.package` overrides the package. If `pkgs.skillnet` is not
available in your package set, use
`inputs.skillnet.packages.${pkgs.system}.skillnet`.
## Storage backends
Postgres is the default calibration backend:
```toml
[database]
url = "postgres://skillnet@localhost/skillnet"
```
SQLite is available by selecting it explicitly:
```toml
[database]
backend = "sqlite"
path = "/home/alice/.local/share/skillnet/multi-phase-plan/calibration.sqlite"
```
The `path` key is optional. Without it, `skillnet` uses
`$skillnet_DATA_DIR/multi-phase-plan/calibration.sqlite`,
`$SKILLNET_DATA_DIR/multi-phase-plan/calibration.sqlite`, or
`$XDG_DATA_HOME/skillnet/multi-phase-plan/calibration.sqlite`.
Select Postgres with an environment variable:
```sh
export SKILLNET_DATABASE_URL='postgres://skillnet@localhost/skillnet'
skillnet calibration migrate
```
Or with `skillnet.toml`:
```toml
[database]
url = "postgres://skillnet@localhost/skillnet"
```
Or through the Home Manager module:
```nix
programs.skillnet = {
enable = true;
package = inputs.skillnet.packages.${pkgs.system}.skillnet;
database = {
backend = "postgres";
urlFile = config.age.secrets.skillnet-pg-url.path;
};
};
```
`--database-url <URL>` overrides both environment and config for one command.
Plain URLs in `programs.skillnet.database.url` are written into the Nix store;
use `programs.skillnet.database.urlFile` with `sops-nix`, `agenix`, or another
secret-backed file for production credentials.
## Quick Start
Inspect configured scopes and current view drift:
```sh
skillnet status
skillnet scope list
```
Materialise derived views and then inspect or regenerate catalog output:
```sh
skillnet sync
skillnet skill list --scope global
skillnet catalog generate
```
`skillnet sync` is dry-run-on-conflict for promotion candidates: a real
view-side directory newer than canonical prints `would promote ...` and exits 2. Rerun with `--apply-promote` to pull that content into canonical. See
[docs/src/commands.md](docs/src/commands.md) for the full flag table.
Calibration commands are available under the dedicated command group:
```sh
skillnet calibration analyze --format table
skillnet calibration proposals --pending
```
## Development
Run the default test suite with:
```sh
cargo test --all-targets
```
Postgres parity tests run when `SKILLNET_TEST_PG_URL` points at a test database:
```sh
export SKILLNET_TEST_PG_URL='postgres://skillnet@localhost/skillnet_test'
cargo test-pg
```
## Configuration Model
`skillnet` keeps canonical skill stores separate from generated agent views:
- `global/` stores the canonical global skills by default.
- Project canonical stores live at each project's `canonical_rel`, defaulting to `.skills`.
- Global and project views such as `.agents/skills` and `.claude/skills` are symlinks generated from canonical stores.
Configuration lives in `$XDG_CONFIG_HOME/skillnet/skillnet.toml`. Catalog
metadata uses `$XDG_CONFIG_HOME/skillnet/skillnet.catalog.toml`. The legacy
cwd fallbacks `./skillnet.toml` and `./skillnet.catalog.toml` still work in
`0.6.x` with a deprecation warning and are scheduled for removal in `0.7.0`.
Run `skillnet config migrate` from the old config directory to centralise
existing files.
## Command Surface
The current top-level commands are:
- `status`
- `completions`
- `sync`
- `view`
- `skill`
- `scope`
- `project`
- `catalog`
- `calibration`
Generate shell completions with:
```sh
skillnet completions bash
skillnet completions zsh
skillnet completions fish
skillnet completions elvish
skillnet completions powershell
```
## Documentation
- Docs: <https://docs.rs/skillnet>
- Source: <https://codeberg.org/caniko/skillnet>
## Release Validation
The release-prep flow validates the repository with:
```sh
simit init flake --check --diff
simit release trust check
simit init ci --platform forgejo --check --diff
nix flake check --keep-going --print-build-logs
cargo fmt --all -- --check
cargo clippy --all-targets --all-features -- --deny warnings
cargo test --all-features
RUSTDOCFLAGS="-D warnings" cargo doc --no-deps --all-features
cargo package --list
cargo publish --dry-run
```
## License
Licensed under either of:
- MIT ([LICENSE-MIT](LICENSE-MIT))
- Apache-2.0 ([LICENSE-APACHE](LICENSE-APACHE))