pleb/plugin-helper
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

init repo for a plugin helper

Browse Source
This commit is contained in:
pleb
2026-06-14 09:14:40 -07:00
commit a9881ebec4
1 changed files with 378 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/DESIGN.md
+378
View File
@@ -0,0 +1,378 @@
# plugin-helper Design
`plugin-helper` is a Python CLI for managing Beat Saber plugins in a mounted Windows BSManager install. It installs from individual GitHub releases, keeps per-game-version plugin selections pinned, records exact filesystem changes, and leaves compatibility judgment visible enough for an agent to help when upstream packaging is inconsistent.
The initial target is the Linux side of `incineroar`, after the Windows partition has been mounted manually. The current Beat Saber instances live under:
```text
/home/pleb/Windows/Users/pleb/BSManager/BSInstances
```
## Goals
- Manage plugins for one BSManager Beat Saber instance at a time, such as `1.40.8`.
- Pull plugin releases directly from configured GitHub repositories.
- Determine candidate updates while respecting the pinned Beat Saber version.
- Support selective updates and explicit pins.
- Resolve dependencies before install, including plugin libraries.
- Install through a dry-run plan first, then apply exactly that plan.
- Record every installed file so uninstall and rollback are deterministic.
- Back up `UserData` separately from plugin installation.
- Package the CLI with a Nix flake so it can be consumed by `brenise.dev/flake.nix` for `incineroar`.
## Non-Goals
- Replacing BSManager as a GUI or Beat Saber instance manager.
- Downloading or downgrading Beat Saber versions.
- Running `nixos-rebuild switch`.
- Mutating the Windows partition unless the user has mounted it and explicitly runs an apply command.
- Treating Nix as the plugin installer. Nix should package `plugin-helper`; the CLI should manage the mutable mounted game tree.
## Core Model
The project should keep four concepts separate:
- **Registry**: checked-in knowledge about known plugins and where releases come from.
- **Lockfile**: the selected plugin versions for a specific Beat Saber version.
- **Install state**: exact files currently installed by `plugin-helper`.
- **Plan**: a generated dry-run artifact describing the next filesystem changes.
This separation lets the script stay deterministic while an agent helps with ambiguous release metadata, compatibility notes, or per-plugin install rules.
## Files and Directories
Recommended repository layout:
```text
plugin-helper/
flake.nix
pyproject.toml
DESIGN.md
README.md
registry/
plugins.toml
locks/
1.40.8.lock.toml
src/plugin_helper/
__init__.py
cli.py
config.py
github.py
instances.py
metadata.py
planner.py
installer.py
state.py
userdata.py
tests/
```
Runtime state should not need to live inside the repository. By default, keep mutable state under an XDG data directory, while allowing explicit paths for easy inspection:
```text
~/.local/state/plugin-helper/
instances/
1.40.8/
installed.json
plans/
downloads/
backups/
```
For early development, a `--state-dir` option is useful so plans and manifests can be kept in the repo while the format settles.
## Registry
The registry describes plugin sources and install behavior. It should be human-editable because many Beat Saber plugins have small packaging differences.
Example:
```toml
[[plugins]]
id = "songcore"
name = "SongCore"
repo = "Kylemc1413/SongCore"
asset_patterns = ["*SongCore*.zip"]
install_strategy = "bsipa-zip"
category = "library"
[plugins.compatibility]
source = "metadata-or-release-notes"
[[plugins.dependencies]]
id = "bs-utils"
constraint = ">=1.0"
required = true
```
Important fields:
- `id`: stable local identifier. Prefer BSIPA metadata ID when known.
- `repo`: GitHub `owner/name`.
- `asset_patterns`: release asset glob patterns, evaluated in order.
- `install_strategy`: known extraction/copy strategy.
- `dependencies`: registry-level fallback when release metadata is missing or incomplete.
- `pin`: optional repository-wide pin, usually better placed in a per-version lockfile.
- `compatibility`: hints for interpreting release metadata.
## Lockfile
Each Beat Saber version should have its own lockfile. This is where selective update and "wait for the ecosystem" behavior lives.
Example:
```toml
beat_saber_version = "1.40.8"
instance = "1.40.8"
[[plugins]]
id = "songcore"
repo = "Kylemc1413/SongCore"
tag = "v4.3.0"
asset = "SongCore-4.3.0.zip"
sha256 = "..."
reason = "Pinned until dependent mods support newer SongCore"
[[plugins]]
id = "some-ui-mod"
repo = "example/some-ui-mod"
tag = "v1.2.1"
asset = "SomeUIMod.dll"
sha256 = "..."
```
The lockfile should be the source of truth for reproducible installs. `check` may propose newer versions, but `apply` should install what is present in a plan generated from the lockfile or from an explicit update command.
## Install State
The install state records what happened on disk. It should not be hand-edited during normal use.
Example:
```json
{
"instance": "1.40.8",
"beatSaberVersion": "1.40.8",
"plugins": {
"songcore": {
"repo": "Kylemc1413/SongCore",
"tag": "v4.3.0",
"asset": "SongCore-4.3.0.zip",
"assetSha256": "...",
"installedAt": "2026-06-14T16:00:00Z",
"files": [
{
"path": "Plugins/SongCore.dll",
"sha256": "...",
"size": 123456
}
]
}
}
}
```
Uninstall should delete only files recorded for that plugin, and should warn before deleting a file whose current hash no longer matches the recorded hash.
## Plans
All mutating operations should have a dry-run plan. The plan is both user-facing and machine-readable.
Plan generation should:
1. Read the target instance and Beat Saber version.
2. Read the registry and lockfile.
3. Query GitHub releases as needed.
4. Select compatible candidate releases.
5. Resolve dependencies.
6. Download release assets into a staging directory.
7. Inspect archives and DLL metadata where possible.
8. Compute target file changes.
9. Write a plan file.
Plan application should:
1. Re-verify downloaded asset hashes.
2. Re-verify the mounted instance path.
3. Back up overwritten files.
4. Copy or extract files exactly as listed in the plan.
5. Write install state atomically.
6. Leave a concise summary of changed files.
## Compatibility
Compatibility should be conservative. A release should be considered compatible only if one of these is true:
- Embedded BSIPA metadata declares the target game version.
- Release notes or asset names explicitly mention the target game version.
- The registry contains a maintained compatibility override.
- The user or agent has reviewed and pinned the release for that Beat Saber version.
When compatibility is ambiguous, `check` should report the candidate as "needs review" instead of proposing an automatic update.
Useful data sources:
- GitHub release tag, title, body, publish date, and assets.
- BSIPA metadata embedded in downloaded DLLs.
- `manifest.json` or similar files inside release archives.
- Existing installed DLL metadata.
- Registry overrides for known edge cases.
bs-manager has a `ModMetadata` shape worth mirroring:
```text
name, id, version, gameVersion, dependsOn, conflictsWith, loadAfter, loadBefore
```
For `plugin-helper`, this metadata should be part of release inspection and dependency solving, not just display.
## Dependency Resolution
Dependency resolution should start simple and explicit:
- Build an available-version set from registry entries and inspected candidate releases.
- Add selected plugins from the lockfile or command line.
- Add transitive dependencies from embedded metadata and registry fallback rules.
- Reject plans with unresolved required dependencies.
- Warn about conflicts.
- Prefer already locked dependency versions unless an update is explicitly requested.
This is closer to a small package solver than bs-manager's current dependency closure. It does not need to be sophisticated at first, but it should keep enough structure to grow into constraint handling.
Initial dependency policy:
- If a dependency is locked, keep it.
- If a dependency is installed but not locked, treat it as unmanaged and warn.
- If a dependency is missing and the registry has it, select the newest compatible release.
- If multiple compatible releases exist, choose the newest stable semver tag.
- If no compatible release is clear, stop and require review.
## Install Strategies
Start with a few explicit strategies instead of generic archive magic:
- `dll-to-plugins`: copy one DLL to `Plugins/`.
- `zip-to-pending`: extract archive under `IPA/Pending/`.
- `bsipa-zip`: install BSIPA-style archive at the Beat Saber root or pending path as appropriate.
- `root-zip`: extract archive relative to the Beat Saber instance root.
- `manual`: plan only; requires a plugin-specific rule before apply.
Every strategy should produce a concrete file list before apply. If an archive cannot be mapped safely, the plan should fail.
## CLI
Proposed commands:
```text
plugin-helper instances
plugin-helper scan --instance 1.40.8
plugin-helper check --instance 1.40.8
plugin-helper plan --instance 1.40.8
plugin-helper plan --instance 1.40.8 --update songcore
plugin-helper apply path/to/plan.json
plugin-helper uninstall --instance 1.40.8 songcore
plugin-helper pin --instance 1.40.8 songcore v4.3.0
plugin-helper backup-userdata --instance 1.40.8
```
Command behavior:
- `instances`: list BSManager instances found under the configured root.
- `scan`: inspect installed plugins and compare against recorded state.
- `check`: query GitHub and report updates, pins, ambiguity, and missing dependencies.
- `plan`: produce a dry-run plan without mutating the game tree.
- `apply`: execute a previously generated plan.
- `uninstall`: remove files recorded in install state.
- `pin`: update the lockfile for a plugin.
- `backup-userdata`: snapshot `UserData` separately.
## UserData Backups
`UserData` backups should be independent from plugin install state because those files change when the game runs, not when plugins are installed.
Initial backup behavior:
- Create timestamped archives under the state directory.
- Include a small manifest with source instance, file count, total size, and hashes.
- Avoid pruning automatically until a retention policy is configured.
- Later, consider incremental backups or `rsync`-style deduplication if archives get large.
## Python Implementation
Python is the recommended language for the first version because this project needs fast iteration, readable plugin-specific rules, and good filesystem/archive ergonomics more than a static binary.
Suggested dependencies:
- `typer` or `click` for CLI.
- `pydantic` for typed config, lockfile, state, and plan schemas.
- `httpx` for GitHub API calls.
- `packaging` for version parsing.
- `tomli-w` for writing TOML if needed.
- Standard library `tomllib`, `zipfile`, `hashlib`, `pathlib`, and `json`.
Keep domain logic in plain Python modules, not inside CLI command functions. The CLI should parse arguments, call services, and print summaries.
## Nix Flake
The flake should package the CLI and expose it as both a package and app:
```nix
{
outputs = { self, nixpkgs }: {
packages.x86_64-linux.default = ...;
apps.x86_64-linux.default = {
type = "app";
program = "${self.packages.x86_64-linux.default}/bin/plugin-helper";
};
};
}
```
Use `buildPythonApplication` once `pyproject.toml` exists. The Nix package should install the tool; it should not encode the mutable Windows mount path or plugin selection.
`incineroar` can then consume it as a flake input and install the CLI into the user or system environment.
## Agent Skill Boundary
An agent skill would be useful after the CLI skeleton exists. The skill should instruct the agent to:
- Read the registry, lockfile, and install state.
- Run `plugin-helper check` and inspect ambiguous releases.
- Use GitHub release notes and downloaded metadata to classify compatibility.
- Edit registry rules when a plugin has unusual packaging.
- Generate a plan, review it, then run `apply` only after the plan is concrete.
- Summarize changed files from the install state.
The skill should not encourage arbitrary manual copying. Manual copying should become a registry rule or an install strategy so future uninstalls remain reliable.
## Safety Rules
- Default all mutating commands to dry-run or require an existing plan.
- Refuse to apply if the BSManager instance path does not exist.
- Refuse to apply if the target path is outside the selected Beat Saber instance.
- Warn before overwriting files not recorded in current state.
- Back up overwritten files before replacement.
- Use atomic writes for lockfiles and state files.
- Never delete files during uninstall unless they are recorded in install state.
- Warn when current file hashes differ from recorded hashes.
## Open Questions
- Should lockfiles live in the repo, XDG state, or both?
- Should registry compatibility overrides be committed, or kept in a local override file?
- Should GitHub API calls require a token to avoid rate limits?
- How should prereleases be handled for mods that publish beta compatibility first?
- Does BSIPA need to be managed by `plugin-helper`, or assumed to be installed by BSManager?
- Should `plugin-helper` optionally consume BeatMods metadata as an advisory source while still downloading from GitHub?
## Initial Milestones
1. Create the Python package, CLI skeleton, and flake.
2. Implement instance discovery and `instances`.
3. Define registry, lockfile, state, and plan schemas.
4. Implement GitHub release discovery and `check`.
5. Implement archive/DLL inspection enough to produce file plans.
6. Implement `plan` and `apply` for `dll-to-plugins` and `zip-to-pending`.
7. Implement uninstall from install state.
8. Add `backup-userdata`.
9. Add an agent skill once the commands and formats are stable.