Knowledge Islands agentic radar
This governance skill maintains a portable, reviewable snapshot of agentic subjects without turning observation into adoption.
Read the agentic-radar standard before assessing or changing the snapshot. The generated rubric publishes its criteria, the radar snapshot holds structured state, and the tracked sources govern refresh inputs.
What each subject is belongs to ki-agentic-systems, whose card the subject names in its card field, such as the Model Context Protocol; the radar never defines it again. The radar does not choose an agent architecture, install integrations, mutate runtime configuration, or treat a protocol version, standards-body incubation, package count, or vendor claim as interoperability proof. ki-model-radar owns models and routes, ki-acquire pulse owns public-signal discovery, capability REFRESH modes own capability-specific policy, and ki-work next owns downstream work capture.
Operating modes
Invoked as help, -h, or ?, emit the generated HELP block and stop. With no recognised mode, emit the same HELP and, only in an interactive session, offer a mode choice.
Mode AUDIT
Run ki repo audit --skill ki-agentic-radar --repo <repo> for deterministic schema, evidence, lifecycle, ownership, and classification checks. Then review judgment criteria in the generated rubric: whether evidence is independent and applicable enough for the recorded maturity and whether architectural terms remain meaningfully distinct.
AUDIT is read-only. Report mechanical findings before judgment gaps and never infer a preferred stance from source count.
Mode CONFORM
Run AUDIT first, then ki repo conform --skill ki-agentic-radar --repo <repo> --dry-run. Only the generated rubric publication is automatically conformable. Changes to radar.toml are authored and guarded because even a syntactic-looking correction can change maturity, stance, movement, or ownership meaning.
Mode EDUCATE
Explain the subject kinds, evidence classes, maturity and implementation dimensions, interoperability states, stance and movement vocabulary, owner and return-trigger requirements, and the adoption boundary. Use examples to distinguish agent loops, branching supervisor trees, graph-orchestrated control flow, knowledge graphs, and provenance graphs. Do not scaffold or mutate a repository.
Mode REFRESH
REFRESH applies only to this skill's canonical files in ki-agent-backpack. From an installed copy, stop and redirect to the harness.
Read the tracked sources, perform a bounded current-evidence review, and reconcile proposed changes against the standard. Preserve uncertainty and counter-evidence. Provider claims alone cannot move a subject inward; Adopt or Trial requires a concrete Knowledge Islands use case and local evidence.
Route material consequences to the owning capability REFRESH mode, a Decision Record, or ki-work next. Never mutate a consumer directly from REFRESH.
Notes
- A community group is not automatically a formal standard, and a versioned protocol is not automatically stable.
- Reference implementations, independent implementations, conformance results, and demonstrated interoperability are separate evidence states.
- Architectural patterns and vendor terms use
not-applicablespecification maturity and must not be labelled formal standards. - Similar graph vocabulary does not collapse control flow, stored knowledge, and evidence lineage into one concept.