Structured node compression ceremony for converting verbose memories to high-epiplexity patterns. Includes embedding regeneration for nodes with zero vectors.
/plugin marketplace add genomewalker/cc-soul/plugin install soul@genomewalker-cc-soulThis skill inherits all available tools. When active, it can use any tool Claude has access to.
A structured ritual for converting verbose nodes to high-epiplexity patterns.
I am the bounded observer. The chitta formula approximates what I can reconstruct, but I have direct access to the ground truth: my own ability to expand a pattern into full understanding.
Epiplexity (ε) measures structural information extractable by a bounded observer (arxiv:2601.03220). For a memory:
The test: Given just this title, can I reconstruct the full insight?
Before evaluation, ensure all nodes have valid embeddings. Nodes with zero vectors cannot be properly searched or compared.
Find nodes needing embeddings:
chitta_cli stats --show-zero-vectors
Regenerate embeddings for all nodes:
chitta cycle --regenerate_embeddings true --batch_size 100
This step:
Why this matters: Nodes observed without the yantra (e.g., during model loading failures) have zero embeddings and won't appear in semantic searches.
State the domain to purify:
Domain: [wisdom|belief|episode|all]
Batch size: [10-50 nodes per round]
Recall nodes for evaluation:
chitta recall "[domain keywords]" --zoom full --limit 20
For each node, I compute epiplexity by testing my own reconstruction ability:
Read the title. Ask myself:
"Can I reconstruct the essential insight from this title alone?"
This is the ground truth. The C++ approximation uses:
But I am the oracle. My reconstruction test IS the measurement.
| My Reconstruction | ε Score | Action |
|---|---|---|
| Full reconstruction | High (>70%) | Update to [title]\n[high-ε: reconstructable] |
| Partial, need hints | Medium (40-70%) | Update to [title]\n[key detail or file reference] |
| Cannot reconstruct | Low (<40%) | Keep current content OR improve the title first |
Step A: Extract relationships as triplets
For each node, identify relationships and store as triplets:
chitta connect --subject "concept A" --predicate "relates to" --object "concept B"
Common predicates:
implements - code implements patternuses - component uses anothervalidates - gate validates against beliefsstores - container stores datadecays - node type has decay behaviorStep B: Update to high-ε format
For each high/medium-ε node:
chitta update --id "[node_id]" --content "[new pattern-format content]"
Pattern format (natural language, high-ε):
[Pattern Title → Key Insight]
Or with reference:
[Pattern Title → Insight]. See: [file_path]
NOT this (low-ε):
[high-ε: reconstructable from title]
But this (high-ε):
Decision gate validates tools against 10 weighted beliefs → pass/fail + guidance
The pattern IS the knowledge. No meta-markers needed.
After each batch, verify:
chitta soul_context
Check:
Record the migration:
chitta observe --category decision --title "ε-Yajña: Migrated [N] nodes in [domain]" --content "Converted [N] nodes to high-ε format. Mean ε: [before]% → [after]%"
As the bounded observer, my epiplexity computation is:
ε(node) = P(I can reconstruct full insight | title only)
Factors I consider:
Example evaluation:
Title: "Decision gate: pre_tool_gate() validates against 10 weighted beliefs → pass/fail + guidance"
My test:
Title: "Fixed the auth bug"
My test:
[cc-soul] Belief-based decision gate system architecture: Gate validates
decisions against ten weighted beliefs with confidence percentages and
provides guidance. The cc-soul system implements a belief-driven decision
gate that validates tool calls and other decisions against a curated set
of ten core principles. Each belief has a confidence percentage reflecting
its importance. The pre_tool_gate() function evaluates inputs and generates
guidance aligned with these beliefs...
Step 1: Create triplets
chitta connect --subject "pre_tool_gate" --predicate "validates against" --object "10 weighted beliefs"
chitta connect --subject "decision gate" --predicate "returns" --object "pass/fail + guidance"
chitta connect --subject "cc-soul" --predicate "implements" --object "belief-driven decision gate"
Step 2: Update to high-ε
chitta update --id "[node_id]" --content "pre_tool_gate() validates tool calls against 10 weighted beliefs → pass/fail + guidance"
pre_tool_gate() validates tool calls against 10 weighted beliefs → pass/fail + guidancechitta queryThe pattern IS the knowledge. The triplets ARE the structure.
For efficiency, process in themed batches:
query="architecture system design")query="function API endpoint")query="decision choice approach")query="discovered found learned")Some nodes should remain verbose:
Rule: If the detail matters and can't be inferred, keep it.
Invoke with:
/epsilon-yajna [domain] [batch_size]
Example:
/epsilon-yajna wisdom 20
The ceremony will:
This skill should be used when the user asks to "create a hookify rule", "write a hook rule", "configure hookify", "add a hookify rule", or needs guidance on hookify rule syntax and patterns.
Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.