# Scentience Skills

Agent skill definitions for olfaction-aware AI — part of the [Scentience](https://scentience.ai)
open olfaction platform. Built to the [agentskills.io](https://agentskills.io) open standard;
compatible with Claude Code, OpenAI Codex, and Google Antigravity.

## Skill Registry

| Skill | Purpose |
|-------|---------|
| [ble-device](./ble-device/SKILL.md) | Connect to Reconnaisscent or Scentinel via BLE; sample or stream raw sensor readings |
| [olfactory-navigation](./olfactory-navigation/SKILL.md) | Plume contact classification and chemical source localization |
| [olfactory-inertial-odometry](./olfactory-inertial-odometry/SKILL.md) | GPS-denied position estimation by fusing olfaction signals with IMU |
| [colip-embeddings](./colip-embeddings/SKILL.md) | Cross-modal retrieval and semantic labeling via COLIP models |

## Decision Guide

| Intent | Use |
|--------|-----|
| Connect to hardware, read sensor data | [ble-device](./ble-device/SKILL.md) |
| I have live OPU readings — where should I go? | [olfactory-navigation](./olfactory-navigation/SKILL.md) |
| Navigate without GPS using smell + motion | [olfactory-inertial-odometry](./olfactory-inertial-odometry/SKILL.md) |
| Semantically label or match a smell episode | [colip-embeddings](./colip-embeddings/SKILL.md) |
| Classify a chemical compound from sensor data | [colip-embeddings](./colip-embeddings/SKILL.md) (`ovl-classifier` model) |

## Typical Workflows

**Real-time robotic navigation:**
`ble-device` → `olfactory-navigation` → motor controller

**GPS-denied localization:**
`ble-device` → `olfactory-inertial-odometry` → path planner

**Semantic labeling pipeline:**
`ble-device` → `colip-embeddings` → natural language annotation

**Full hybrid system:**
`ble-device` → `olfactory-navigation` + `olfactory-inertial-odometry` + `colip-embeddings`

## Design Principles

- **Olfaction as first-class modality** — chemical sensing receives the same structural
  rigor as vision or audio; no scalar collapsing of temporal signals
- **Temporal fidelity** — all skills preserve signal history, intermittency, and trend
  context across the full reading window
- **Calibrated uncertainty** — every output carries a `confidence` field; no skill
  returns conclusions without explicit bounds
- **Control ≠ semantics** — navigation decisions (`olfactory-navigation`,
  `olfactory-inertial-odometry`) and semantic interpretation (`colip-embeddings`) are
  deliberately separate skills with separate output schemas
- **SDK-grounded** — all examples use the official Scentience SDK; no pseudocode

## SDK

| Language | Package | Install |
|----------|---------|---------|
| Python | [scentience](https://pypi.org/project/scentience/) | `pip install scentience` |
| JavaScript | [scentience](https://npmjs.com/package/scentience) | `npm install scentience` |
| Rust | [scentience](https://crates.io/crates/scentience) | `cargo add scentience` |
| C++ | [scentience/2.0.0](https://scentience.jfrog.io/ui/packages/conan:%2F%2Fscentience/2.0.0) | Conan |

API docs: [scentience.github.io/docs-api](https://scentience.github.io/docs-api/) ·
API keys: [dashboard.scentience.ai](https://dashboard.scentience.ai)