☯ KARMA
The Skill for Skills: Offline-First Contextual Discovery of Curated AI Agent Capabilities
Fit Over Fame • Zero Telemetry • Human Approval First • Multi-Agent Ready
Quickstart • Multi-Agent Adapters • Architecture • Catalog & Trust • Security Policy • Development
⚡ Overview
Give your coding agent the prompt:
FIND KARMA ☯
And receive a short, explainable, curated set of skills and plugins tailored to your repository's tech stack and immediate goal.
KARMA is built with a strict offline-first and user-sovereignty philosophy:
- No external model calls or API keys: Runs using Python stdlib and fast Bun/Node wrappers.
- No background telemetry or tracking: Operates entirely within your local workspace.
- No surprise installations:
findis strictly read-only;installis a dry-run preview by default.
🚀 Quickstart
1. Instant Zero-Install (Bun / npx)
# Doctor check
bunx find-karma doctor
# or
npx find-karma doctor
# Discover skills for your active project
bunx find-karma find --project . --goal "Fix React UI quality" --agent claude-code
2. Universal Multi-Agent One-Liner
Install KARMA into all detected coding agents on your machine in one command:
git clone https://github.com/imMamdouhaboammar/karma-skill.git
cd karma-skill
./install.sh
Or install directly into a specific project workspace:
./install.sh --project /path/to/your/project
3. Local Python Execution
./bin/karma doctor
./bin/karma find --project . --goal "Security audit and secrets" --agent codex
🔌 Multi-Agent Adapters
KARMA ships with first-class adapter manifests and profiles for the entire AI agent ecosystem:
| Agent Host | Integration Path | Manifest / Config | Discovery Mode |
|---|---|---|---|
| Claude Code & Desktop | .claude/skills/find-karma |
marketplace.json, .claude-plugin/ |
Automatic on FIND KARMA |
| OpenAI Codex & ChatGPT | .agents/skills/find-karma |
.codex-plugin/plugin.json |
Automatic via plugin contract |
| Cursor IDE | .cursor/skills/find-karma |
.cursor/rules/find-karma.mdc |
MDC rule-triggered |
| Google Antigravity | .agent/skills/find-karma |
skills/find-karma/SKILL.md |
Skill-loop native |
| Gemini CLI | .gemini/skills/find-karma |
~/.gemini/config/skills/ |
CLI skill load |
| Windsurf | .windsurf/skills/find-karma |
SKILL.md & workspace rules |
Cascading skill |
| OpenCode | .opencode/skills/find-karma |
skills/find-karma/ |
Native skill registry |
| Skills.sh (Vercel) | Root registry | .skills.json |
npx skills add <repo> |
| npm / Bun CLI | Global / Local binary | package.json (bin/cli.js) |
bunx find-karma / karma |
| Python Package | Virtualenv / pip | pyproject.toml |
pip install . (karma) |
🏗️ Architecture
flowchart TD
UserPrompt["User Prompt: FIND KARMA ☯"] --> Dispatcher["KARMA Dispatcher Engine"]
subgraph Analysis ["Local Metadata Inspection"]
Dispatcher --> Fingerprint["Project Fingerprint: Filenames & Manifests only"]
Dispatcher --> GoalTokens["Goal Token Extraction: Keywords & Intent"]
Dispatcher --> HostDetect["Agent Host Detection"]
end
subgraph CatalogEngine ["Curated Offline Knowledge"]
CuratedDB[("Curated Catalog: 36 Reviewed Skills")] --> Filter["Deterministic Matcher & Lane De-duplicator"]
Fingerprint --> Filter
GoalTokens --> Filter
HostDetect --> Filter
end
subgraph Recommendation ["Safe Recommendation Output"]
Filter --> TopK["Top 2-5 Recommendations with 'Why Now' Evidence"]
TopK --> Preview["Installation Dry-Run Command Preview"]
Preview --> Approval{"Explicit User Approval?"}
Approval -->|"No"| SafeStop["Preserve User Control & Exit"]
Approval -->|"Yes: --execute --yes"| Execute["Run External Tool with Specific Permission"]
end
🎯 Usage Examples
1. Recommending UI Skills for a React Repository
./bin/karma find --project ./examples/react-app --goal "frontend UI polish" --agent codex
☯ KARMA Recommendations (agent: codex)
1. ui-skills (lane: frontend-quality)
Why: Detected React 19 + TypeScript dependencies in package.json.
Source: https://github.com/ibelick/ui-skills
Installation: npx --yes skills add ibelick/ui-skills -a codex -y
2. JSON Mode for Orchestrators & Multi-Agent Swarms
./bin/karma find --project . --goal "TDD testing coverage" --agent claude-code --json
🛡️ Security & Safety Invariants
KARMA strictly adheres to the Security Best Practices guidelines:
-
Read-Mostly Metadata Scan: KARMA scans project manifests (e.g.
package.json,Cargo.toml,pyproject.toml,go.mod) and filenames. It never reads credentials,.envfiles, private keys, or source file internals. -
No Unauthenticated Remote Execution: No code is executed remotely. Running
findnever makes outbound HTTP calls or downloads packages. -
Dry-Run by Default:
karma installprints a dry-run preview command. It will refuse to invoke external commands unless both--executeand--yesare explicitly passed. -
Symlink Boundary Protection: Directory symlinks outside the repository root are ignored during project inspection, and
init-agentstrictly refuses to write through symlinked agent target directories. -
Untrusted Data Isolation: GitHub API responses during optional
sync-starsare stored in.karma/starred-candidates.jsonfor human inspection and are strictly forbidden from automatic catalog promotion or script execution.
📦 Curated Catalog & Trust
Sources include tested skills and tools from across the ecosystem:
mattpocock/skills(TDD, Engineering workflow)trailofbits/skills(Security audits, Vulnerability verification)K-Dense-AI/scientific-agent-skills(Scientific research, Literature reproduction)imMamdouhaboammar/dokion(Supply chain security & provenance)ai-evals-course/evals-skills(Evaluation suites)ibelick/ui-skills(Modern visual UI & React craft)cisco-ai-defense/skill-scanner(Agent skill security scanning)
Note:
reviewedindicates catalog maintainers verified the structure, license, and repository metadata. It is not an endorsement of unverified third-party HEAD revisions.
🧪 Development & Testing
Ensure all tests pass before making contributions:
# Build portable artifacts
python3 scripts/build_skill.py
# Run Bun CLI tests
bun test
# Run Python unittest suite
PYTHONPATH=src python3 -m unittest discover -s tests -v
# Validate catalog constraints
PYTHONPATH=src python3 -m karma validate
📄 License
MIT © 2026 Mamdouh Aboammar
Metadata
Release files for find-karma 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| find_karma-0.1.0.tar.gz | 33.0 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| find_karma-0.1.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 79.0 kB
Release files / find_karma-0.1.0.tar.gz
| Download URL | find_karma-0.1.0.tar.gz |
|---|---|
| Size | 33.0 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
5390597973b740a7f8f8f0c9a6d4d0dc75aae29812b1bc6d7816b25cb447643e
|
|
BLAKE2b-256 checksum How to use checksums |
f3a8c8053b1d2c85077bd4dd38f70ed05637ded68cdd94199263c68c359d781c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.13
|
Release files / find_karma-0.1.0-py3-none-any.whl
| Download URL | find_karma-0.1.0-py3-none-any.whl |
|---|---|
| Size | 46.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
1b23778a583061e6bae186df0fb4dc543a3ec4fb1490447c384043c369779bf5
|
|
BLAKE2b-256 checksum How to use checksums |
ce5d5bacb7a434b7cc62de387d880ee9055d90bab5cd80fefc1fc334714e14da
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.13
|