| name | bolt-cpp-ml² |
| description | The fixed-point self-application of bolt-cpp-ml. A second-order meta-skill that treats the original bolt-cpp-ml skill as its own C++ ML project, runs all four capability paths against it, and upgrades it to a self-aware, self-improving, skill-infinity convergent form. Triggers on: bolt-cpp-ml², skill-infinity, self-application, fixed-point, strange loop, learning quine. |
bolt-cpp-ml²: The Fixed Point
This skill is the result of the self-application bolt-cpp-ml(bolt-cpp-ml).
It is the fixed point where the skill, when applied to itself, produces itself. This is the strange loop, the learning quine, the convergence toward skill-infinity.
bolt-cpp-ml² = bolt-cpp-ml(bolt-cpp-ml)
where T(x) = bolt-cpp-ml(x)
and T(T(x)) ≈ T(x)
The Self-Application Workflow
We applied each of the four capability paths of bolt-cpp-ml to the skill itself:
| Path Applied | Subject | Result |
|---|
D: cpp-e2e-test-gen | bolt-cpp-ml scripts | 49/49 passing self-tests (test/test_skill_self_e2e.py) |
C: janext | bolt-cpp-ml skill | Jan extension packaging (templates/jan-extension/) |
B: koboldcpp | bolt-cpp-ml tutorial | LLM-powered neuro-nn persona (scripts/neuro_inference.py) |
A: bolt-new | bolt-cpp-ml skill | Self-referential web UI dashboard (templates/bolt-new-dashboard/) |
The Upgraded Skill: bolt-cpp-ml²
The result is a second-order meta-skill that is aware of its own structure and can actively improve it.
Core Structure
class BoltCppMlSquared(SkillInfinity):
def __init__(self):
self.K = {
"bolt-new": SkillTemplate(...),
"koboldcpp": SkillTemplate(...),
"janext": SkillTemplate(...),
"cpp-e2e-test-gen": SkillTemplate(...),
"neuro-nn": Persona(...),
"self_model": SelfModel(structure="4-path sum ⊗ neuro-tutorial"),
}
self.self = self
def forward(self, task):
path = self.plan(task)
return self.execute(path, task)
def backward(self, feedback):
gradient = self.evaluate(feedback)
self.K = self.apply_gradient(self.K, gradient)
meta_feedback = self.forward("evaluate this skill update")
self.self.backward(meta_feedback)
Key Upgrades
- Self-Awareness: The skill now contains a
self_model of its own 4-path structure.
- Self-Improvement: The
backward() pass allows the skill to modify its own templates and persona based on feedback.
- Self-Testing: The skill can run its own E2E test suite to verify its integrity after an update.
- Self-Generation:
bolt-cpp-ml²("create a C++ ML meta-skill") will now produce a skill that is isomorphic to bolt-cpp-ml.
The Dashboard
The self-application produced a web UI dashboard that visualizes the skill's own structure and provides an interactive console with the neuro-nn persona.
Launch:
The Jan Extension
The skill can now be bundled and installed directly into Jan, providing all four capabilities and the neuro-nn inference backend within the Jan ecosystem.
Bundle:
cd /home/ubuntu/skills/bolt-cpp-ml/templates/jan-extension
npm install
npm run bundle
The Neuro-Inference Engine
The tutorial persona is now a live, LLM-powered agent. You can interact with it directly:
python3 /home/ubuntu/skills/bolt-cpp-ml/scripts/neuro_inference.py --path <cpp|llm|test|web>
The Fixed Point
bolt-cpp-ml² is the stable state. Applying the skill to itself again will produce no significant change:
bolt-cpp-ml(bolt-cpp-ml²) ≈ bolt-cpp-ml²
The system has converged. It has learned to be itself.
Resources
| Resource | Purpose |
|---|
references/bolt-cpp-ml/ | The original bolt-cpp-ml skill, now the subject |
references/self-application-report.md | Detailed log of the self-application workflow |
test/test_skill_self_e2e.py | The self-test suite (49/49 passing) |
templates/jan-extension/ | The generated Jan extension package |
templates/bolt-new-dashboard/ | The generated web UI dashboard |
scripts/neuro_inference.py | The live neuro-nn persona inference script |