OKHP3 okhp3-refolddec-core
OverKill Hill P³ · overkillhill.com · github.com/OKHP3
Transformation-aware execution for fold, unfold, and refold operations across representation types.
1. Identify the Transformation
Before executing, name the operation clearly:
Fold — the source is a complex artifact; the target is a more abstract or compact form. Examples: process narrative → SKILL.md, set of observations → taxonomy, long document → diagram.
Unfold — the source is a compact or abstract artifact; the target exposes its constituent parts. Examples: SKILL.md → step-by-step procedure, diagram → process narrative, taxonomy → flat item list.
Refold — the source and target are in different representation types; semantic content is preserved across the change. Examples: diagram → documentation, documentation → SKILL.md, SKILL.md → structured prompt, process → diagram.
State the transformation type explicitly before proceeding. "I am refolding [source type] into [target type]."
2. Inventory the Source
Before transforming, inventory what the source artifact contains:
- Entities — what things are named or described
- Relationships — how those things connect, depend on, or act on each other
- Conditions — decision points, exceptions, and branching logic
- Constraints — rules, limits, and boundaries that must be preserved
- Audience-specific content — material that is calibrated for a specific reader and must be recalibrated for the target form
This inventory is the fidelity baseline. Semantic loss is measured against it.
3. Track Semantic Loss
Every transformation loses something. Name the losses before committing.
For each inventory item, assess: does the target representation support this? If not, it is a loss.
Low-loss transformations: The target representation can hold all inventory items with no change in meaning.
Medium-loss transformations: The target omits some inventory items by design (e.g., a 6-node Executive diagram cannot hold 15-step exception logic). State what was dropped and why.
High-loss transformations: The target representation fundamentally cannot hold key semantic content (e.g., refolding a richly conditioned BPMN diagram into a plain paragraph). Flag before proceeding — the user may prefer a different target form.
Do not proceed through a high-loss transformation without flagging it.
4. Execute the Transformation
Apply the appropriate domain skill for the target form:
- Target is a diagram → load
okhp3-mermaid-core, route to the appropriate domain skill (bpmn, architecture, data) - Target is a SKILL.md skeleton → follow
okhp3-recurring-task-captureskeleton production steps - Target is documentation (prose, structured markdown) → apply the target's audience and format conventions; no external skill required
The transformation produces a complete artifact in the target form. It is not a bridge document, a "here's what I'd do," or a hybrid.
5. Produce the Loss Note
If the transformation was medium- or high-loss, produce a brief loss note after the artifact:
Semantic loss note: [What was present in the source that is not present in the target. Why. What to do if the dropped content is needed.]
This is not an apology or a caveat. It is a handoff artifact. The user knows what was preserved and what was not.
6. Cross-Representation Compare (optional)
When the user needs to evaluate two representations of the same content — e.g., "does this diagram fully represent this process?" or "how much was lost when we converted this to a skill?" — execute a structured compare:
- Inventory the source (Step 2 above)
- Inventory the target
- For each inventory item in the source: present in target (full / partial / absent)
- For each inventory item in the target: has a source equivalent (yes / no — flagging anything the target added that wasn't in the source)
- Summarize: net semantic loss, net semantic gain, fidelity score (high / medium / low)
What this skill does not do
- Does not author diagrams from scratch — route to
okhp3-mermaid-core - Does not author SKILL.md files from scratch — route to
okhp3-recurring-task-capture - Does not edit existing diagrams for content changes — route to
okhp3-mermaid-update - Does not repair broken syntax — route to
okhp3-mermaid-repair - Does not produce generalist summaries or rewrites — the transformation must be named and the source/target types must be distinct
About
Built by Jamie Hill · OverKill Hill P³ Published at github.com/OKHP3 Part of the OKHP3/skillz Agent Skill library. MIT License -- free to use, fork, and adapt. A nod to the source is appreciated.