repository·main·Indexed 23 days ago
https://github.com/rohitg00/skillkitA package manager for AI agent skills that allows developers to install, translate, and manage skills across 46 different AI agents, such as Claude Code, Cursor, and Copilot. SkillKit provides a CLI for project-aware skill recommendations, security scanning for prompt and command injection, session memory management, and the ability to sync project context across multiple agents.
The TUI provides several specialized screens for managing your agent skills and workflows:
The mesh network supports different security profiles depending on your environment:
| Level | Discovery | Transport | Auth | Use Case |
|---|---|---|---|---|
development | open | none | none | Local dev |
signed | signed | optional | optional | Trusted LAN |
secure (default) | signed | required | required | Production |
strict | trusted-only | required | mTLS | High security |
Design can be applied at different granularities depending on the scope of the task:
SkillKit uses a central SKILL.md (Universal Format) as the intermediary. The translation flow follows this path:
To implement a new adapter, the interface requires two methods:
getConfigPath(): Returns the skill directory.generateConfig(): Converts the content into the target format.Skills display trust badges based on their source and a quality analysis performed by the TrustScorer.
The score is a weighted composite of four dimensions:
The task-decomposition skill is used when a project, feature, or research task feels too large to manage. The core principle is to break work down into small, atomic units that are independently completable, verifiable, and should ideally take no more than a few hours to finish.
This skill is triggered by phrases such as:
The extension follows a specific pipeline to convert web content into agent-ready skills:
activeTab permission. No content script injection is required.skillkit.sh/api/save-skill or agenstskills.com/api/save-skill).SKILL.md and uses the chrome.downloads API to save the file to your disk.Note on Privacy: The extension only sends the page URL to the server. It does not send page content, browsing history, or personal data. Selection-based saves are performed entirely locally in the browser.
SkillKit's memory system follows an Observation → Learning Pipeline to convert AI session patterns into reusable skills:
To minimize token usage, the system uses a 3-layer retrieval model. It starts with the lightest layer and fetches deeper layers only as needed based on relevance and your token budget:
| Layer | Content | ~Tokens |
|---|---|---|
| Index | Titles, tags, timestamps | 50-100 |
| Timeline | Context, excerpts, activity | ~200 |
| Details | Full content, metadata | 500-1000 |
Memories in @skillkit/mcp-memory are organized by categories and tiers to facilitate efficient retrieval and lifecycle management.
fact: Factual informationdecision: Decisions madepreference: User preferencespattern: Observed patternsinsight: Derived insightsreasoning: Reasoning chainswarm: Recent, frequently accessed memorieslong: Consolidated long-term memoriesLifecycle: Memories are automatically promoted from warm to long based on access count, reinforcement score, and age.
The TUI uses specific visual cues to communicate skill quality and status.
Skills are graded based on their score:
●: Enabled○: Disabled⟁: Claude Code◫: Cursor◎: Codex✦: Gemini⬡: OpenCodeWhen using getRelatedSkills, you can filter by the following relation types:
similar: Skills that share tags and functionality.complementary: Skills that work well together.dependency: Skills required by other skills.alternative: Different approaches to the same problem.