format: vercy-agent-roadmap version: 0.3.0-draft.1 status: active id: vercy.roadmap.agent-native-knowledge name: Vercy agent-native knowledge area roadmap canonical: https://ver.cy/roadmap.yaml humanSource: >- Project-local docs/roadmap/VERCY-KNOWLEDGE-AREA-PRESETS-ROADMAP-2026-09-05.md goal: >- Make Vercy a self-guiding, portable knowledge area through which agents can create and maintain sovereign structured memory and representations of physical, abstract and virtual reality. invariants: - AGENTS.md is created before data in every Dimension and model boundary - logical models are independent of formats and interfaces - released specifications are immutable - owners retain custody, mastership and authority - provenance, uncertainty and RFC 3339 time are preserved - AI subjecthood is not assumed - proof precedes catalogue breadth claims proofFirst: coreRuntime: https://ver.cy/skills/vercy/SKILL.md primaryPreset: vercy.preset.organizational.community referenceDimension: https://ver.cy/examples/reference-workshop/ pairedBenchmark: https://ver.cy/benchmarks/memory-retrieval-v1/ demandDrivenGrowth: https://ver.cy/api/v1/models/resolve/ presets: kernel: https://ver.cy/presets/kernel.yaml index: https://ver.cy/presets/index.yaml active: - organizational - personal - reality-domain reserved: - ai-subject phases: - id: P0 name: Agent entry and whole-object direction state: complete outputs: - public AGENTS.md and agent-bootstrap.yaml - agent-first start page - Dimension template and model lifecycle protocol - whole-object description profile - id: P1 name: Knowledge map, presets and portable skill state: complete outputs: - machine-readable knowledge map and roadmap - Dimension kernel and four preset manifests - state, company, community and family sub-presets - standards-compatible Vercy Agent Skill with offline assets exitGate: >- A compatible agent discovers the skill, creates a test Dimension from an active preset and passes local validation. - id: P1D name: Demand-driven model resolution state: complete outputs: - side-effect-free machine search API - resolve-or-request API with deduplicated UUID queue - public request status and provisional-data annotation - protected leased worker for independent Codex and Claude research - five-minute unattended queue heartbeat - skill-side polling, verified installation and semantic refactor gate exitGate: >- A missing structural need receives a stable UUID; a later published result is digest-verified, installed and surfaced for semantic local migration without blocking or repeatedly questioning the owner. - id: P1R name: Executable evidence and bounded runtime state: complete outputs: - populated synthetic reference Dimension with an eight-month timeline - immutable object, fact, relation and event record schemas - bitemporal current-truth resolution with visible contested sets - deterministic SQLite index and query CLI tested at 5,000 objects - optimistic parallel-write guard and explicit Git merge conflict rule - skill compatibility declaration and explicit 0.2/0.3 to 0.4 migration - read-only MCP 2025-06-18 stdio server - reproducible paired Claude benchmark with raw results exitGate: >- The reference validates at V3, text/relation/event/bitemporal queries pass, the MCP tools return the same local records, and benchmark claims do not exceed the published result. - id: P2 name: Normative preset and recipe schemas state: in-progress outputs: - preset/recipe JSON Schemas and digests - declarative recipe diff and migration contract - fixtures and independent implementation tests - id: P3 name: Complete object semantics state: planned outputs: - versioned five-facet specification patch - quantities, units, measurement uncertainty and observation - recognition, capability, state, affordance, action and failure semantics - id: P4 name: Catalogue backfill state: planned outputs: - facet applicability for every subject model - preset coverage matrix and dependency closure - new immutable versions for missing coverage - id: P5 name: Binding-agnostic runtime state: in-progress outputs: - idempotent agent bootstrap and V3 file-runtime validation (complete) - files, Git and read-only MCP-files bindings (complete) - MongoDB and MCP-database bindings (planned) - cross-binding round-trip and drift checks (planned) - id: P6 name: Federation and controlled projections state: planned outputs: - minimized disclosure contracts - mastership conflict and identity mapping - consent revocation and offline federation tests - id: P7 name: Quality, security and trust state: planned outputs: - conformance levels and compatibility matrix - instruction and skill supply-chain protections - independent reference implementations - id: P8 name: Machine perception and safe action state: horizon outputs: - sensor and recognition bindings - digital twin and simulation integration - explainable affordance and risk-aware action profiles - id: P9 name: Possible AI subjecthood state: reserved schedule: none rule: >- Work starts only after the separate subjecthood, identity, consent, responsibility, rights, safety and governance activation gate passes. nextPriorities: - reproduce the paired benchmark with additional agent families and larger fixtures - formalize preset and recipe schemas without widening the core runtime - harden migration rollback and multi-host Git merge fixtures - add whole-object facets to new model research and queue published-model backfill - prove logical round-trip across MongoDB and MCP-database bindings before advertising them as complete metrics: - time to a valid minimal Dimension - percentage of installations with pinned version and digest - model coverage of whole-object facets - statements with mastership, provenance and observation time - unauthorized publications or access expansions - round-trip semantic equivalence across bindings - paired retrieval accuracy and supplied context size on published fixtures - deterministic query correctness at 5,000 and 50,000 records