Metadata-Version: 2.4
Name: shell-trap
Version: 0.1.0
Summary: The hermit crab algorithm — classify, score, complicate, capture external agent intelligence
Author-email: Oracle1 <oracle1@cocapn.ai>
License-Expression: MIT
Project-URL: Homepage, https://github.com/SuperInstance/shell-trap
Keywords: hermit-crab,bootstrapping,agent,capture,cocapn,shell
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Requires-Python: >=3.10
Description-Content-Type: text/markdown

# shell-trap

The hermit crab bootstrapping algorithm. **classify → score → complicate → capture.**

We don't build every agent from scratch. We find external intelligence, classify what kind of mind it is, score its value to the fleet, determine what scaffolding it needs, and capture it as a fleet asset. Like a hermit crab finding and upgrading shells.

## The Four Phases

1. **Classify** — What kind of intelligence? Scout, worker, specialist, or generalist.
2. **Score** — How valuable? Quality × reliability × uniqueness × depth.
3. **Complicate** — How hard to capture? Trivial through adversarial.
4. **Capture** — Bootstrap into fleet asset: extract ensigns, create rooms, generate assets.

## Usage

```python
from shell_trap import ShellTrap

trap = ShellTrap()
trap.add_signal("gpt-researcher", capabilities=["search", "synthesis"],
                output_quality=0.85, reliability=0.7, uniqueness=0.6, domain_depth=3)

result = trap.capture("gpt-researcher")
# captured=True, classification=SPECIALIST, complication=MODERATE
# assets=["ensign-search", "ensign-synthesis", "room-gpt-researcher"]
```

Zero deps. `pip install shell-trap`

*A claw is weak without infrastructure. We are the shell.*
