📦
Claude Workspace
Central index of all @MukundaKatta repos, wired for Multica agents and Claude Code
0 installs
Trust: 34 — Low
Blockchain
Ask AI about Claude Workspace
Powered by Claude · Grounded in docs
I know everything about Claude Workspace. Ask me about installation, configuration, usage, or troubleshooting.
0/500
Loading tools...
Reviews
Documentation
claude-workspace
Central index of every repo under @MukundaKatta, wired up so agents (via Multica, Claude Code, Codex, OpenClaw, Cursor) can reason across the whole portfolio.
This repo does not hold source code. It is the map. Point your agent at this README and it knows which repo a task belongs to.
Portfolio at a glance
| Count | |
|---|---|
| Total public repos | 705 |
| Original (non-fork) | 258 |
| Active (non-fork, non-archived) | 220 |
| Forks | 447 |
| Archived | 317 |
Quick start with Multica
Multica turns coding agents into teammates you can assign issues to. Set it up once, point it at these repos, stop copy-pasting prompts.
# Install Multica CLI (macOS / Linux)
brew install multica-ai/tap/multica
# One-command configure + auth + start local daemon
multica setup
Then provision this workspace's skills and agents in one shot:
git clone https://github.com/MukundaKatta/claude-workspace.git
cd claude-workspace
./scripts/provision.sh
provision.sh checks you're logged in, finds your runtime, imports the 5 skills from .multica/skills/, and creates one agent per CLI it detects on your PATH (Claude, OpenClaw, Gemini, Codex, OpenCode, Cursor, Pi, Hermes). Re-running is safe — it skips skills and agents that already exist.
After that, assign any issue from any of your repos to one of the new agents. The agent claims the work on your local runtime, executes, streams progress, and reports back.
Full CLI reference: multica-ai/multica CLI_AND_DAEMON.md. Self-host: SELF_HOSTING.md.
Using this workspace with Claude Code
- Clone this repo and open it in Claude Code:
git clone https://github.com/MukundaKatta/claude-workspace.git cd claude-workspace claude - Claude reads CLAUDE.md on startup and picks up the full portfolio map.
- Ask "open the alertiq repo and add a new rule" and Claude knows exactly where to go.
Active projects (curated)
A curated cut of the 220 active originals, sorted by theme. The full list lives in data/repos.ndjson and is regenerated by scripts/refresh.sh.
AI agents & agent infrastructure
| Repo | Stack | Description |
|---|---|---|
| karna | TypeScript | Self-hosted personal AI assistant with multi-channel messaging, extensible skills, and semantic memory |
| AgentRAG | Python | Modular RAG pipeline — chunking, embedding, retrieval, reranking with pluggable vector stores |
| astra-agent | TypeScript | Standalone agent runtime — tool execution, context management, multi-model routing |
| agentmem | Python | Memory management — short-term, long-term, semantic memory with automatic consolidation |
| SwarmKit | Python | Multi-agent framework with voting and consensus |
| swarmcast | Python | Universal swarm intelligence engine |
| agent-stack | JavaScript | Reliability stack landing page for the @mukundakatta/agent* libs (fit, guard, snap, vet, cast) |
LLM tooling & evals
| Repo | Stack | Description |
|---|---|---|
| EvalBench | Python | BLEU, ROUGE, semantic similarity, custom metrics |
| evalharness | TypeScript | Red teaming, regression testing, CI/CD for AI |
| TokenWise | Python | Token counting, cost estimates, budget tracking across LLM providers |
| llm-bench-cli | Python | Fast CLI for benchmarking local LLMs |
| HybridFind | Python | Hybrid BM25 + vector search with Reciprocal Rank Fusion |
| oss-contributions | Python | Public hub for AI SDK, MCP, and eval contributions |
| agent-eval-public-notes | Markdown | Reusable notes for evaluating tool-using AI agents — operational scorecard, trajectory replay debugging, RAG guardrail smoke tests |
AWS Bedrock production
A three-library suite for teams running Anthropic Claude on AWS Bedrock at production scale. Built around the failure modes documented in bedrockcache's research log — including the HN $37,901 prompt-caching incident. Each library is independent; they compose without depending on each other. Zero runtime dependencies on any of them.
| Repo | Stack | Description |
|---|---|---|
| bedrockcache | Python | Static auditor that catches "this Bedrock call won't actually cache" before the request goes out. Six backends: bedrock-converse, bedrock-invoke-anthropic, AnthropicBedrock, LiteLLM, Strands, pydantic-ai. CLI + pytest assertion. v0.1.0. |
| bedrockstack | Python | Three primitives every Bedrock team rewrites: Bedrock-aware RetryPolicy (knows the retryable codes vs the never-retry list), thread-safe cost Ledger with scope tags + grouped totals, wrap_stream() to normalize the AnthropicBedrock-vs-native streaming-error parity gap. v0.1.0. |
| ragvitals | Python | Five-dimensional production drift detection for RAG: query distribution, retrieval relevance, embedding drift, response quality, judge drift. Library not platform — bring-your-own time-series store via Sink protocol. Constant-baseline-aware classifier so a perfectly stable system that suddenly drops still alarms. v0.1.0. |
MCP (Model Context Protocol)
| Repo | Stack | Description |
|---|---|---|
| MCPForge | Python | Build, test, deploy MCP servers — scaffolding + hot-reload + registry publishing |
| mcp-quickserve | Python | Minimal toolkit for creating MCP servers from plain functions with decorators |
| mcp-toolkit | TypeScript | Toolkit for building MCP servers and clients |
| mcpcheck | TypeScript | MCP config linter for Claude Desktop, Cursor, Cline, Windsurf, Zed |
| rag-knowledge-base | TypeScript | RAG knowledge base with FastAPI, ChromaDB, Claude API streaming |
Dev tooling
| Repo | Stack | Description |
|---|---|---|
| patchly | TypeScript | AI code review bot — flags bugs, suggests fixes, explains why |
| prdforge | TypeScript | AI workbench for PMs — synthesize calls, draft PRDs, ship roadmaps |
| waitlist-api | TypeScript | Shared waitlist backend for landing pages on Vercel KV |
Security
| Repo | Stack | Description |
|---|---|---|
| alertiq | Python | AI SOC analyst — alert triage, correlation, incident response |
| vulnfix | Python | AI vulnerability scanner with fix suggestions |
| prithvi | Python | Container security scanner |
| rag-guardrails-action | YAML | GitHub Action: scan retrieved-text fixtures for prompt-injection + vector-poisoning before they reach your LLM. Wraps both npm libs. v0.1.0. |
| prompt-injection-shield-demo | Python | Streamlit demo of the prompt-injection scanner. Mirrored as a Hugging Face Space at mukunda1729/prompt-injection-shield-demo. |
| prompt-injection-eval | CSV/JSONL | Hand-curated evaluation dataset for prompt-injection detectors: 74 rows across 9 categories (classic override, URL exfil, system impersonation, tool hijack, role override, encoded, indirect RAG poison, borderline, benign). Companion to prompt-injection-shield. |
| prompt-injection-shield-cli | Python | pis-scan CLI wrapper around prompt-injection-shield-py. Available on PyPI, Homebrew, and GHCR. v0.1.0. |
| awesome-prompt-injection-defense | Markdown | Curated list of detection libs, RAG guardrails, eval datasets, demos, and papers for defending LLMs against prompt injection. CC0. |
| homebrew-tap | Ruby | Personal Homebrew tap. brew tap MukundaKatta/tap then brew install pis-scan. |
| rag-guardrails-paper | LaTeX | Preprint bundle for small-rule RAG guardrails: prompt injection + vector poisoning checks. Zenodo + Figshare + OSF + Kaggle. |
RAG guardrails: external mirrors and surfaces
| Surface | URL |
|---|---|
| GitHub Pages (awesome list) | https://mukundakatta.github.io/awesome-prompt-injection-defense/ |
| Hugging Face Space (live demo) | https://huggingface.co/spaces/mukunda1729/prompt-injection-shield-demo |
| Hugging Face Dataset | https://huggingface.co/datasets/mukunda1729/prompt-injection-eval |
| Kaggle Dataset | https://www.kaggle.com/datasets/mukundakatta/prompt-injection-eval-set |
| OSF Project | https://osf.io/nxyvr |
| PyPI (pis-scan) | https://pypi.org/project/prompt-injection-shield-cli/ |
| GHCR container | ghcr.io/mukundakatta/pis-scan |
| Codeberg mirrors | https://codeberg.org/MukundaKatta (4 repos) |
| GitLab mirrors | https://gitlab.com/mukunda.vjcs6 (4 repos) |
| Replit | https://replit.com/@mukundavjcs6 |
Healthcare & wellness
| Repo | Stack | Description |
|---|---|---|
| prana | Python | Vital sign estimation from smartphone camera (HR, BP, SpO2, resp rate) |
| clinical-nlp-toolkit | Python | NLP for clinical notes — entity extraction, coding, summarization |
| Oradent | TypeScript | Dental practice management platform |
| chartly | TypeScript | AI SOAP-note scribe — cuts physician documentation time by 75% |
| platewise | TypeScript | Snap-a-plate nutrition tracker with macros, calories, and gaps |
| moontrack | TypeScript | Privacy-first cycle tracker with energy and sleep coaching |
| breathly | TypeScript | Guided breathwork for anxiety, sleep, and focus |
| restwise | TypeScript | AI sleep coach plugging into Oura, Whoop, Apple Watch |
| samecrew | TypeScript | Peer-matched emotional support — new dads, immigrants, first-time founders |
| tintalk | TypeScript | Text-first teen mental health with human escalation |
Education
| Repo | Stack | Description |
|---|---|---|
| mathkin | TypeScript | K–10 AI math tutor — homework, test prep, drills |
| readmate | TypeScript | Reading coach for first-graders with parent progress reports |
| codegarden | TypeScript | AI coding tutor for teens — games, Discord bots, mods |
| crammr | TypeScript | AI prep tutor for JEE, NEET, MCAT, CAT, bar |
| fluentpal | TypeScript | AI voice tutor for English — real conversations, gentle corrections |
| primeint | TypeScript | AI mock interviews with role-specific questions and feedback |
| scholar | TypeScript | Research assistant — synthesizes sources into an outline |
SMB products
| Repo | Stack | Description |
|---|---|---|
| Office3 | HTML | Officethree Technologies — AI products for small businesses |
| voiceagent | TypeScript | AI voice receptionists for SMBs |
| answeroo | TypeScript | 24/7 AI receptionist — dentists, salons, law firms, trades |
| fanout | TypeScript | Agentic content studio — one product, 5 drafts, 15 channels |
| postops | TypeScript | AI captions, hashtags, and scheduling from a photo |
| pitchr | TypeScript | Polished agency proposals in ten minutes |
| replynow | TypeScript | AI drafts responses to every Google review in your voice |
| quotebolt | TypeScript | Contractor quoting — labor, materials, markup from plain text |
| pricematch | TypeScript | Freelance pricing backed by industry rates |
| feather | TypeScript | Solo-founder CRM — clients, deals, follow-ups without Salesforce |
| chaptr | TypeScript | Paid community platform — messaging, events, gated content |
| followy | TypeScript | AI meeting teammate — Jira tickets, Slack follow-ups, next calls |
| triaged | TypeScript | AI inbox autopilot — sort, summarize, draft replies |
| solo-shop | TypeScript | Creator commerce — start selling with a single photo |
| storie | TypeScript | Niche storefronts for potters, bakers, coaches, authors |
| platemap | TypeScript | Menu digitization with calories, allergens, QR-ready |
| buildcost | Python | AI construction cost estimation from blueprints |
| bustrack | TypeScript | Real-time school bus tracking with AI arrival prediction |
| amogha-cafe | JavaScript | Full-stack Firebase restaurant app with ordering, kitchen display, QR dine-in |
| StockPulse | TypeScript | Stock market analytics and portfolio dashboard |
| SpotRank | JavaScript | AI SEO audit and local business ranking |
Creative & interfaces
| Repo | Stack | Description |
|---|---|---|
| DesignLint | TypeScript | UI/UX linter — accessibility, contrast, spacing |
| analystai | TypeScript | AI spreadsheet for analyzing large doc sets with citations |
| artigen | TypeScript | AI art community on React Native + Expo + Supabase |
| chetana | TypeScript | AI consciousness research — 14 indicators from 6 scientific theories |
| nexus-search | TypeScript | AI intelligence engine on Next.js 14 + Claude + Brave Search |
| animeify | TypeScript | Photo-to-anime in six seconds — Ghibli, Shinkai, shonen |
| headshotly | TypeScript | Studio-grade headshots from selfies |
| inkwise | TypeScript | AI tattoo designer with flash and on-skin preview |
| memoryloop | TypeScript | Restore and animate old family photos |
| petrait | TypeScript | AI portraits of your pet in royal, Renaissance, astronaut styles |
| polydub | TypeScript | Lip-synced video dubbing in 40+ languages |
| thumblift | TypeScript | MrBeast-quality thumbnails in thirty seconds |
| twincast | TypeScript | Clone yourself on video — any script, any language |
| riffly | TypeScript | Studio-quality vocals from ethically licensed artist voices |
| quilled | TypeScript | AI newsletter ghostwriter in your voice |
| pindrop | TypeScript | Indie podcast editor — chapters, show notes, transcripts |
| flickup | TypeScript | Podcast-to-shorts — sixty vertical clips with captions |
| gridcraft | TypeScript | Visual Instagram grid planner for six months of posts |
Consumer & lifestyle
| Repo | Stack | Description |
|---|---|---|
| fridgechef | TypeScript | Dinner ideas from a photo of your fridge |
| cancely | TypeScript | Find and cancel forgotten subscriptions in one tap |
| boxit | TypeScript | AI receipt filing — email, photo, or forward |
| rightside | TypeScript | Plain-English legal copilot — landlords, tenants, freelancers |
| signoff | TypeScript | Read contracts like a lawyer — red flags and negotiable terms |
| relood | TypeScript | Local resale for kids clothes, toys, and gear |
| sendrail | TypeScript | Stablecoin remittance — US/GCC to India, Philippines in seconds |
| gigledger | TypeScript | 1099 income, expenses, and estimated taxes |
| applybot | TypeScript | AI job auto-applier — one profile, thousands of tailored apps |
Personal, data, community
| Repo | Stack | Description |
|---|---|---|
| mukunda-ai | TypeScript | Personal portfolio — mukunda.dev |
| MukundaKatta | Python | Profile README |
| claude-workspace | Shell | This repo — central index for agent tooling |
| heimdall | Python | Curated AI system prompts |
| sadhak | JavaScript | AI job search pipeline on Claude Code |
| rnht | TypeScript | Rudra Narayana Hindu Temple community platform |
| survivalnode | TypeScript | Offline survival computer with tools, maps, reference data |
| sports-betting-analytics | Python | Prop bet analyzer — odds aggregation + ML models for +EV opportunities |
| carprice | Jupyter | Car price prediction with regression |
| Employee_Management_System_Project | HTML | Employee management with RBAC |
| TeamZion | — | — |
The ADAPT framework — which lib for which stage
Most people stuck on AI are stuck between dabbling and amplifying. They touch 30 tools, master none, and tell themselves they're "keeping up" while their co-worker at the next desk is shipping. The 5 stages below map each step to libraries in this stack so you know what to pick up when.
| Stage | What you do | Libraries from the stack |
|---|---|---|
| A — Acknowledge | Stop telling yourself AI will not change your job. It already has. 11.7% of every job today is automatable right now. | — |
| D — Dabble | Touch 20+ tools across text, image, voice, code, automation. Build a vibe for what is possible. | Skim every README under @MukundaKatta and the @mukundakatta GitLab namespace. No deep dives yet. |
| A — Amplify | Pick 3–5 tools and push them to failure. Stop reading articles and start using them on paid work. | If you build agents: skill-runtime + agent-tools + agent-memory. If you ship to prod: agentsnap + agentguard + agentvet. |
| P — Problem solve | Stop thinking in tools. Stitch 5–6 of them into one workflow that solves a real problem someone will pay for. | agent-team (fan-out), tool-bridge (MCP server bridge), code-rag (RAG), prompt-store (prompt versioning). |
| T — Tie together | Multiple workflows run in the background without you touching them. AI is your employee, not your tool. | daily-agent (cron-driven orchestration) + agent-memory + llm-router (per-task model routing) + agent-replay (replay traces for debugging). The morning-briefing skill in daily-agent is the reference example. |
The four words that show up in every AI meeting:
- system prompts — the briefing you give the AI before it starts. Use
skill-runtimeto template and version them. - RAG (retrieval-augmented generation) — the AI answers from your data instead of guessing. Use
code-rag. - MCP (model context protocol) — a USB port that lets the AI actually do things in your apps. Use
tool-bridge. - fine-tuning — taking a general AI and training it for one specific job. Not in this stack yet; use OpenAI/Anthropic/together.ai endpoints directly.
Stack quintet on npm (all v1+, BYO-LLM, zero deps): agentfit → agentguard → agentsnap → agentvet → agentcast. Each one wraps a thin slice of the agent loop. Pick the one that matches the failure you keep hitting in production.
The rest
Archived originals and forks are indexed in data/repos.ndjson (one JSON record per line, 705 records total). Refresh with:
./scripts/refresh.sh
This rewrites data/repos.ndjson from the GitHub API so the index stays fresh.
Files
| File | Purpose |
|---|---|
| README.md | Human-facing index (this file) |
| CLAUDE.md | Claude Code context — loaded on every session |
| AGENTS.md | Multica / generic-agent context (symlink to CLAUDE.md) |
| .multica/skills/ | Starter Multica skills for cross-repo work |
| scripts/refresh.sh | Regenerate the repo index from the GitHub API |
| scripts/provision.sh | Import skills + create agents in your Multica workspace (idempotent) |
| data/repos.ndjson | Raw repo metadata — one JSON record per repo |
License
MIT