probelisted

# Probe-driven validation (DOC-164 slice 8+) The agentic ontology's MVP acceptance is **probe-driven**, not coverage-%-driven (per `adrs/drafts/research/agentic-code-ontology/PROBES.md`). Two rounds the maintainer authors and the skill executes: - **Type 4 — Adversarial probes (per refresh).** Three maintainer-authored "capability negation" probes (capabilities that *don't* exist — synonym-swap-with-negation, cross-product-fabrication, or capability-negation). The ontology should answer "no" / return empty. ≥2 of 3 must PASS for the refresh to be accepted. - **Type 6 — Implicit-ADR confirmation (per refresh, but lower frequency in practice — quarterly).** Maintainer privately writes 5 ADRs they know the codebase embodies. The skill compares against `lineage/{repo}/implicit-adrs.md`'s top-10 candidates. ≥3 of 5 must appear (semantically equivalent) for the implicit-ADR feature to be supported. <3 PASS = block ontology MVP acceptance. This skill is **maintainer-driven** — the implementer cannot author probes (that defeats the adversarial-round design per LSN-013-class case-law: implementer-authored probes only test what the implementer expected the ontology to handle). ## Argument forms | Form | Behaviour | |---|---| | `/probe --adversarial [<repo>]` | Capture 3 adversarial probes the maintainer types in. Run each against the ontology (`concepts.yaml` + `understanding/*.md`). Report ≥2/3 PASS or FAIL. | | `/probe --implicit-adrs [<repo>]` | Capture 5 maintainer-written imp