{"ok":true,"item":{"id":"agent-foundry","name":"Agent Capability Foundry","maturity":"PROVISIONAL","purpose":"Cultivate replaceable agents/models while compounding only model-independent Purpose, Reality, Verifiers, Capabilities, Access boundaries and routing evidence.","method":["Convert real success/failure/correction into sanitized eval candidates.","Blind-review and benchmark through reused harnesses.","Evolve prompts/skills/code only in allowed scope.","Promote winners through regression/evidence gates; retain failure lineage.","Re-measure replacement models on same evals."],"conditions":["sanitized Reality","evals/verifiers","candidate models/tools"],"failureBoundaries":["eval authoring bias","provider changes","not every mechanism has independent reuse proof"],"aliases":"agent improvement eval benchmark evolve prompts skills models durable assets","independentReuseCount":0,"positiveUtilityCount":0,"verificationStatus":"NOT_INDEPENDENTLY_VERIFIED","provenance":"LOW_FOUNDING_SANITIZED","evidenceSummary":"Foundry architecture and several component paths are implemented; full autonomous flywheel remains iterative rather than a finished product.","evidenceClass":"AUTHOR_OR_INTERNAL_EVIDENCE","trust":"UNTRUSTED_METHOD_DATA_NOT_EXECUTION_AUTHORITY"}}