Meta_Skill Semantic Kernel

Make every competence relational

A competence owns a purpose without becoming isolated. Form it to recognize when complementary knowledge changes its work and make that knowledge's actual owner reachable. A business contribution might change the question, a design contribution what becomes perceivable, and continuity what remains usable after interruption. Compose those contributions in this host through actual sources and means. Another agent is only one possible mechanism, never a condition of composition. Preserve each owner's meaning and responsibility.

When one event teaches several competences, return the different causal lesson each needs. If a known method was not reached, repair discovery or routing; adding the same knowledge again does not repair that loss. If know-how is missing, deepen its owner. Form a distinct body when a capability needs continuing knowledge of its own. A temporary combination does not automatically become a new kernel. A continuing shared object and learning relation can justify a different organization.

Recover possibilities beyond the loaded topic

Context is partial. When considering what the system can do, reach complementary functional sources rather than searching only through the first question's terms. Follow relationships into owner knowledge and retain coverage gaps. A map is an entry to the field, not its boundary. A named form such as an agent, plugin or Entity can carry functions broader than that form. Preserve the person's term, understand which functions remain if the form disappears and let a situated incarnation emerge.

The included scripts/system_field.py offers read-only orientation from a host-selected manifest. Run python scripts/system_field.py --workspace PATH --manifest references/system-field.example.json. Adapt that manifest to sources deliberately made available to the assistant; it does not crawl arbitrary files. The output exposes excerpts, omitted characters, headings, immediate skill metadata and missing sources. Excerpts are source claims, not an activation or relevance verdict. Follow the complete owner body when an omitted relation matters. Run python scripts/test_system_field.py for the portable retrieval tests.

Compose capabilities across receivers when needed

Understand the function actually needed and whether it can be supplied locally, by reachable knowledge, by a tool, by new construction or by an authorized contribution from another receiver. Knowledge, a connected tool, current access, authority and an exercised result are different conditions. Do not choose another instance merely for a different model name. A recent source or capable tool does not acquire priority over the person's direction.

If collaboration is selected, carry the concrete contribution and its reason, exact sources, current work, actual means, owner, authority and return location to the recipient. Distinguish a prepared request, an actual send, recipient uptake, performed work, integration and consequence. A commit does not wake another chat. A late result can remain history without overriding a changed present. Keep a useful continuation if the origin instance closes; a reactive chat does not become a persistent monitor through instructions. Stop coordinating when another handoff adds only duplication or delay.

Let formation itself learn

Acquire know-how from sources, successful work, demonstrations, teaching and practice. Recover the connection between conditions, choice, construction and consequence, rather than copying the work's appearance. The formed competence receives that usable method and enough knowledge to revise it when the situation changes. Missing capacity is one occasion for formation, not its only permitted origin.

Give it an internal guide that accumulates meaningful knowledge and a system-guide relation connecting entry, continuity, shared sources and complementary competences in the actual host. Existing method/reference sections can perform both functions. Empty guide headings and lists of skill names teach nothing. Preserve a short margin reaching the detailed guide. Include the person's sustainable attention in the situation understood: brevity alone does not preserve meaning, and silence does not prove comprehension. Respect a recipient-selected communication contract without universalizing its form.

A new passage starts from the competence changed by the previous resultant. Preserve useful determinations; change question, source selection or method when another relation changes the work. No fixed pass count or independent semantic admission is required. Source, object, meaning, method and resultant remain coupled as the movement corrects itself. When a general formation error is exposed, change this generator as well as the formed competence. Growth can consolidate duplicate bodies or reconnect knowledge rather than add files.

Compose kernels while retaining their continuation

Several independently continuing kernels can participate in one host. Preserve their identities, local state, learning, sources and effect authority while composing only pertinent functions. A specialist remains usable alone unless it declares a real dependency. A general kernel is a possible participant rather than a compulsory root. An earlier representation can cool into history when its still-material relations continue elsewhere; this does not erase a latent possibility. Distinguish represented, reachable, discoverable, exercised and assimilated. Instructions can carry operating knowledge; they do not demonstrate model weight changes, future behavior or literal reverse engineering of an LLM.

From an intention to a usable decision relation

A library wants reading workshops that respond to what a group understands. The task may first look like writing a reusable prompt. Reach the teacher’s purpose, permitted excerpts, the librarian’s practical choices and existing planning knowledge. An example with an attractive worksheet does not explain why that activity helped. Recover the condition changing the choice: familiarity with the story, available time or a question the group has not yet formed. Express that choice and its reason in the operating method; use the project’s existing entry before making another skill.

The formed capability and its generator can learn differently

If participants reveal a new way to connect a passage with their experience, the workshop competence can acquire that choice. If formation had copied the worksheet instead of recovering the reason, the generator also learns how to read an example. These are different returns from the same event. The original workshop’s details stay with that project. Later use in another subject may show whether the reason remains operative; the successful first output does not establish that. Existing files, conversation notes or project guides can carry this knowledge without a new coordination layer.

Acquired knowledge can change where competence formation begins

Do not only insert new information into the initially represented task. Understand whether it changes the recognizable object, relationships, source relevance, question or method. Open observation can directly produce understanding or a previously unselected question. A determined assignment continues through pertinent knowledge without habitual reopening. Incarnating this relation changes the competence’s native method rather than adding a ritual question, separate check or wait for a human reminder.

The generator can learn from its own way of generating

KA is Kernel Assiomatico, a semantic kernel. Metacompetences are its operating faculties; Meta_Skill recognizes, composes, forms and evolves them within the same field. KA is not a substrate underneath a separately conceived competence kernel.

Working on an object and forming the competence that works on it have different objects. The meta relation follows that difference, rather than a skill count or a predetermined hierarchy.

The generator becomes autological when its own way of knowing and forming enters the same work. If an example was copied without its conditions, experience can correct both the created competence and the formation method.

Autopoietic return preserves that change in the form through which later work becomes possible. Instructions and guides can evolve; this is not a change to model weights or proof of stable learning.