Skip to content

implementation-status.md

Latest commit

 

History

History
98 lines (89 loc) · 9.83 KB

implementation-status.md

File metadata and controls

98 lines (89 loc) · 9.83 KB
Raw

Implementation Status

Completed

  • Created the monorepo layout for CLI/core, desktop app, and Rust crates.
  • Retired the legacy Node CLI/core packages after migrating their product behavior into Rust.
  • Installed Rust stable with rustup.
  • Migrated the core scan, import, symlink deploy, SQLite indexing, and GitHub URL parsing paths into Rust crates.
  • Replaced the temporary system sqlite3 shell-out with rusqlite parameterized writes.
  • Installed rustfmt and verified Rust formatting.
  • Switched the Tauri bridge from spawning the Node CLI to calling Rust crates directly.
  • Implemented SKILL.md frontmatter parsing.
  • Implemented recursive skill scanning.
  • Implemented user and remote import storage.
  • Implemented symlink deployment with overwrite protection.
  • Implemented the versioned Rust runtime-profile registry for .agents/skills, .codex/skills, .claude/skills, .cursor/skills, and exact custom SKILL.md roots, including schema-v6 workspace backfill, structured frontmatter preservation, compatibility previews, stale-preview deployment confirmation, CLI/Tauri parity, and profile-aware desktop workspaces.
  • Implemented GitHub URL normalization for repository, tree, blob, raw, and contents API URLs, including sanitized standalone repository-root skill snapshots.
  • Implemented a first CLI surface for the planned commands.
  • Added a Tauri + React desktop shell with scan and path bridge commands.
  • Implemented Rust/Tauri user-skills Git sync for the shared ~/.skillbox/user-skills repository, including Settings-managed remote configuration, per-skill dirty status, desktop commit review with diff preview, generated Conventional Commit messages, and selected-file commits.
  • Implemented explicit inbound user-skills Git synchronization on main through Check remote -> Review incoming changes -> Apply fast-forward, with separate worktree/relation state, stale-preview rejection, untrusted-tree validation, deployed deletion/rename blockers, backup refs, and independent Git/SQLite recovery auditing. This implementation is newer than the shipped v0.6.1 release and remains unreleased.
  • Implemented Rust/Tauri/CLI remote skill update status checks, Dashboard status refresh, last-checked timestamps, and configurable 5-minute auto refresh.
  • Implemented GitHub-only remote source search/binding, immutable remote version listing, all-file diff preview, update/rollback apply, and permanent operation logging in Rust core, Rust CLI, Tauri commands, and desktop review dialogs.
  • Implemented network-backed GitHub install preview/apply in Rust core, Rust CLI, and desktop UI, including first-install diff review, preview identity checks, version snapshots, current symlink updates, source metadata, optional CLI deploy, and legacy CLI aliases.
  • Added compatibility coverage for Node MVP SQLite stores, including legacy operations migration and explicit timestamp writes for old skills/deployments tables.
  • Added shared desktop diff rendering and remote skill workflow normalization helpers for source binding and version change previews.
  • Implemented SQLite-backed workspace registry for global and project-local skills roots, including .codex/skills, .agents/skills, .claude/skills, scan-time auto registration, imported skill counts, preview-confirmed single-root project initialization, manual add/forget, Rust CLI compatibility commands, Tauri commands, and a searchable desktop Workspaces page with type filters and per-workspace skill review/import.
  • Implemented import records and import revert in Rust core, Rust CLI, Tauri commands, and Skill Detail UI, including backup restoration, conservative legacy reconciliation, multi-workspace blocking, and warning/danger confirmation states.
  • Implemented Rust-owned Import Review groups with stable content/status variant identities, full-snapshot-equivalent locations, location-level User/Remote advice with explicit mixed-suggestion confirmation, explicit selection for materially different same-name variants, group-level Calls/search/tab counts, and one-primary-only import enforcement that preserves existing revert guarantees.
  • Added signed macOS app update checks and user-confirmed install/restart through the Tauri updater plugin, plus release workflow assets for updater archives, signatures, and latest.json.
  • Added daily macOS updater metadata checks with a SQLite-backed successful-result cache, a sidebar Update reminder, one-click metadata recheck plus signed install/restart, and retry-safe pending updates without automatic downloads.
  • Added ordered, transactional Rust SQLite migrations, consistent pre-migration backups for existing databases, schema version tracking, and integrity validation.
  • Moved dashboard favorites and user-edited tags from browser-only storage into SQLite, with a one-time desktop migration for existing local metadata.
  • Added a read-only Doctor workflow in Rust core, Rust CLI, Tauri, and desktop Settings for checking schema/integrity, managed skill layouts, remote current links, deployments, workspaces, import backups, and stale metadata.
  • Extended operation auditing across direct/reviewed imports, deploy/undeploy, skill type changes, workspace add/forget, user-skills Git remote/sync, and usage hook injection, including failed attempts.
  • Implemented reviewed skill deletion across Rust core, CLI, Tauri, and Skill Detail UI, including all-workspace symlink removal, active-import and foreign-target blockers, preview identity checks, transactional active-state cleanup, and retained deletion backups. Single-workspace removal continues through the deployment picker and shared undeploy_skill core.
  • Implemented evidence-aware local Skill Usage Rankings across Rust core, CLI, Tauri, and desktop. schema v7 persists confirmed/inferred/reference plus bounded provenance, upgrades duplicate events only toward stronger evidence, and idempotently rebuilds skill_usage_stats from confirmed + inferred Calls without requiring a rescan. Codex structured per-turn skill carriers and Cursor transcript assistant Read events are inferred Calls; Cursor Reads deduplicate once per transcript user turn and skill and can retain safe historical-missing lexical evidence without granting filesystem authority. Stop hooks and Claude Code native Skill attribution are confirmed; Cursor state attachments and public usage-record events are references; Cursor ReadFile candidates remain diagnostic-only. Dashboard, Workspaces, History, Skill Detail, and Rankings keep Calls separate from History references. Coverage returns evidence totals, time ranges, overlapping provenance source counts, scan/backfill aggregates, and an aggregate-only usage-audit; it does not expose prompt/chat/tool bodies. Codex provider-native run totals remain unavailable, so Calls are explicitly a known local undercount. Sync histories is user-triggered, idempotent, and can recover or upgrade evidence while preserving successful provider results when another provider fails.
  • Added Rust crate scaffolding for the planned Tauri/Rust architecture.
  • Verified the desktop shell in browser preview at http://127.0.0.1:1420/.

Next Implementation Targets

The 0.6 implementation and release qualification are complete. SkillBox v0.6.1 is the shipped runtime-profile and evidence-aware usage release. The next milestones follow the versioned evolution path:

0.7 — Safe Sync, Deployment, And Recovery

  • Reviewed inbound user-skills Git is implemented on main but remains unreleased after v0.6.1. It separates clean/dirty worktree state from unknown/synced/ahead/behind/diverged/remote-only/no-remote-branch relation, then uses Check remote -> Review incoming changes -> Apply fast-forward.
  • The inbound implementation validates the complete remote skill tree, binds a preview_id, discloses deployed updates, blocks deployed deletes/renames, creates an internal backup ref, and transactionally reconciles the user-skill index with Git compensation on failure.
  • Inbound remains explicit and origin/main-only. It does not run on startup, select individual skills, or automatically merge, rebase, reset, stash, force-push, or resolve conflicts.
  • Add copy-snapshot deployment as an explicit, compatibility-checked alternative without weakening the current symlink protections.
  • Strengthen backup inspection, restore previews, and recovery auditing.
  • Keep runtime profiles, schema-v6 workspace migration, schema-v7 evidence classification, and Calls/reference semantics backward compatible.
  • Keep native non-SKILL.md formats behind the future adapter boundary.

Later Milestones

  • Plan Git-backed Skill Collections for v0.8.0, after the v0.7.0 inbound-sync release. A collection is a canonical Git repository/worktree source plus reviewed ref/HEAD; child SKILL.md directories remain independently selectable and deployable. The planned phases are local repository grouping, persistence and child relationships, one-fetch GitHub multi-skill preview/apply, and commit-consistent collection update/rollback.
  • Collection scans must remain read-only. Apply must revalidate repository identity, HEAD/ref, and tree snapshot; execute no hooks, submodules, filters, scripts, or arbitrary shell; and preserve existing path, overwrite, duplicate-name, backup, and recovery protections. No collection phase is currently implemented.
  • Add FTS-backed search across skills, operations, and usage history.
  • Add remote source provenance and trust classification without treating popularity as verification.
  • Add copy-snapshot deployment and stronger restore/audit workflows.
  • Complete product hardening, release-candidate qualification, and the explicit 1.0 promotion gates defined in the roadmap.

When implementation changes milestone status, scope, or ordering, update this file and docs/roadmap.md in the same change set.