Revert the vite.config.ts changes from e93adaa that changed the dev
server port from 3000 to 8181 and added local dev features (env loading,
serve-skill-md plugin, proxy headers). These were local development
configs that broke CI by changing the port away from 3000 which the
Makefile health check expects.
list.ts:
- Remove locked universal section, all agents are now equal
- Users manually select which agents to list from
- No default selections, cleaner UX
check.ts:
- Add agent filtering via --agent flag or interactive selection
- Add status filtering via --status flag or interactive selection
- Default to showing only OK + Missing (not Orphaned)
- Interactive prompts for scope, agents, and statuses
- Show agent names and scope prefix in output
- Smart defaults reduce noise from orphaned skills
- list.ts: use isUniversalForScope for dynamic grouping based on scope
- list.ts: group output by skill name to avoid duplicate path listings
- uninstall.ts: use isUniversalForScope for dynamic grouping
- uninstall.ts: add printUninstallResults for consistent formatted output
- .npmignore: exclude development files from npm package
- package.json: bump version to 1.1.1
- update.ts: sort skill selection list alphabetically
- check.ts: sort results by status then name, sort agent names in locations
- install.ts: sort agent names in buildAgentSummary, sort --list output
- sync.ts: sort discovered skills by name before display
Ensures consistent alphabetical ordering across all CLI commands for
better user experience and predictability.
Group agents that share the same install path into a single line.
When more than 5 agents share a path, show first 5 names + count.
Sort agent names alphabetically within each group.
Extract buildInstallResultLines utility function to DRY up
duplicate output logic in installFromRegistry and installFromGit.
- Sort universal and non-universal agent lists by name in selectAgentsInteractive
- Exclude agents with showInUniversalList===false from locked section
- Replace spinner.stop() with spinner.succeed() before interactive prompts
to prevent 'Fetching' text from appearing as the step title
- Also fix spinner.stop -> spinner.succeed for 'Installation complete'
- Move scope selection before agent selection in both installFromRegistry
and installFromGit flows
- Use isUniversalForScope in selectAgentsInteractive so that the Universal
locked section correctly reflects the actual scope:
- Project: agents with skillsDir='.agents/skills' (Cline, Codex, Cursor, etc.)
- Global: agents with globalSkillsDir='.agents/skills' (Cline, Warp only)
- Clean up unused getUniversalAgents/getNonUniversalAgents imports
- Add isUniversalForScope(agent, isGlobal) and getAgentTargetDir(agent, isGlobal) to agent-detector.ts
- An agent is 'universal' when its target install dir equals the canonical .agents/skills directory
- This fixes incorrect symlink skipping for agents like Codex/Cursor whose globalSkillsDir
differs from .agents/skills (e.g. .codex/skills, .cursor/skills)
- Update installer.ts to use dynamic check via optional AgentInfo parameter
- Update install.ts to use getAgentTargetDir and pass agent info to installSkill
- Update uninstall.ts to use isUniversalForScope for correct path resolution
- Clean up unused isUniversalAgent import in list.ts
- Stop spinner before runInteractiveSearch to prevent terminal UI conflict
(same issue as version selection p.select)
- Make spinner messages more specific: include skill name, version, path
instead of generic "Fetching"/"Downloading"/"Extracting"
The ora spinner was still running during p.select() interactive prompt,
causing the spinner to continuously overwrite the terminal and block
the version selection UI. The user would see "Fetching..." stuck until
pressing Enter.
Stop spinner before p.select() and restart after selection completes.
- Replace all `npx clawhub` references with `npx motovis-skillhub` / `skillhub`
- Update namespace separator from `--` to `/` in install commands
- Change env var CLAWHUB_REGISTRY to SKILLHUB_REGISTRY
- Add vite dev server plugin to serve /registry/skill.md
- Update i18n strings (zh/en): ClawHub CLI → SkillHub CLI
- Update skill.md/skill.md.template with new CLI commands
- Update quick-start.tsx syntax highlighting and env config
- Update e2e test assertions
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- install: use actualSlug instead of raw slug in addToLock to prevent
namespace doubling (vision2group/vision2group/slug)
- update: remove hardcoded 'node dist/cli.mjs' fallback, rely on
process.argv[1] which is always reliable
- download: use finished() to ensure stream write completes before
proceeding
- undici does not auto-follow HTTP redirects, manually follow 302/307/308
to MinIO presigned URL before piping response body
- Commander.js parses --skill-version as skillVersion (camelCase), not
skill-version (kebab-case), fix option access in install and download
commands
- Use /api/v1/skills (compat API) when query is empty to get full stats
- Map sort options: hot=rating, newest=newest, downloads=downloads
- Remove vim-style k/j navigation to fix 'k' key input issue
- Show '(default)' suffix when namespace is global (user didn't specify --namespace)
- Display 'Status: Published' on success instead of 'undefined'
- Keep output clean and informative for user feedback
* feat(skill): add UPLOADED status for PRIVATE skill lifecycle
## Summary
- Add UPLOADED status for PRIVATE skills after security scan passes
- PRIVATE skill owners can test before confirming publish or submitting for review
- Rerelease now follows visibility rules (PRIVATE→UPLOADED, PUBLIC→PENDING_REVIEW)
- Auto-withdraw changes status to UPLOADED (not DRAFT) to keep versions visible
## Changes
- SkillVersionStatus: Add UPLOADED enum value
- SkillPublishService: PRIVATE skills go to UPLOADED after scan
- SecurityScanService: Visibility-based status transition after scan
- SkillGovernanceService: Withdraw→UPLOADED, delete allows UPLOADED
- SkillQueryService: Include UPLOADED in version list filters
- SkillReviewSubmitService: New service for submit-review and confirm-publish
- SkillLifecycleController: Add submit-review and confirm-publish endpoints
- Frontend: Add buttons, dialogs, and hooks for new operations
## Workflow
- PRIVATE: Publish → SCANNING → UPLOADED → confirm-publish → PUBLISHED
- PUBLIC: Publish → SCANNING → PENDING_REVIEW → PUBLISHED
* feat(review): add backward compatibility for DRAFT status
Support both DRAFT (legacy) and UPLOADED (new flow) status in:
- SkillReviewSubmitService.submitForReview
- SkillReviewSubmitService.confirmPublish
- ReviewService.submitReview (both overloads)
This ensures existing data with DRAFT status continues to work
with the new visibility-based workflow introduced in OSS-02.
* feat(publish): allow warning-confirmed pre-publish checks\n\nFixes #287
* fix(i18n): add missing register validation translation keys
The registration form uses i18n keys like register.usernameInvalid,
register.passwordTooShort etc. but they were never defined in the
locale files, causing E2E tests to fail because the raw key strings
were displayed instead of human-readable messages.
* fix(storage): defer S3 bucket verification until first access
* test(storage): cover deferred S3 bucket verification
* fix(runtime): widen backend container healthcheck window
* fix(runtime): widen backend container healthcheck window
* fix(test): use ddl-auto=create to prevent cross-context table drops
Multiple @SpringBootTest classes with different @MockBean configs
cause separate Spring contexts sharing the same H2 in-memory database.
With create-drop, one context's shutdown drops tables needed by another,
causing "Table not found (this database is empty)" errors.
* fix(test): widen awaitIndexedDocument timeout to 15s
CI runners are resource-constrained and async search indexing may not
complete within the previous 5-second window, causing flaky failures.
