Skip to content

component-map.md

Latest commit

 

History

History
113 lines (89 loc) · 8.52 KB

component-map.md

File metadata and controls

113 lines (89 loc) · 8.52 KB

AgentOps Component Map

Status: routing map for validation-centered DDD + hexagonal architecture. Purpose: line up product direction, bounded contexts, ports/adapters, and open bead posture before new work is shaped. Use with: Intent-to-Loop Hexagon, Skill Ports and Adapters, and Bounded Contexts.

AgentOps narrows to one product direction:

Repo-local autonomous code validation: an in-session loop, verification membrane, and context compiler that prove agent output before autonomy expands.

Everything else is either an adapter that helps that route, a substrate concern outside the product core, or backlog sprawl to defer or trim.

Routing Rule

Before creating a skill, workflow, CLI command, gate, or bead, route it through this map:

  1. Name the component and bounded context.
  2. Name the port behavior, not the tool.
  3. Name concrete adapters separately.
  4. Keep the work only if it proves code or agent output, advances the golden route, measures the corpus delta, hardens the validation membrane, or preserves the repo-local corpus.
  5. Defer or trim anything that only expands surfaces, duplicates the loop, or belongs to an external product/domain.

Components

Component Bounded context Owns Ports / objects Product posture
Context Corpus BC1 Corpus .agents/, .ao/wiki, retrieval, capture, citation, learning promotion, corpus durability CorpusReaderPort, CorpusWriterPort, ContextCompilerPort, CitationPort, FindingCompilerPort, Finding, Learning, ContextPacket Supporting core. This is the durable asset and candidate moat, but the quality claim is unproven until measured.
Verification Membrane BC2 Validation gates, CI/local gate verdicts, council/pre-mortem/validate, safety policy, evidence binding, release readiness GateRunnerPort, CIStatusPort, SafetyPolicyPort, ClaimEvidenceBinderPort, CriterionVerdict, Evidence Product center and correctness floor. Keep it fail-closed; autonomy expands only when this stays trustworthy.
Work Graph and Loop Engine BC3 Loop BDD intent, br beads, work selection, slice plans, RPI, evolve, closeout, convergence shape_intent, persist_intent, plan_slices, execute_slice, execute_wave, record_evidence, steer_goal, LoopReaderPort, LoopWriterPort, CloseoutPort, ConvergenceCheckPort, WorkItem, Slice, Wave Core route. One loop body, one inner tick, one in-session driver.
Skill / Claim Factory BC4 Factory skill/workflow admission, skill quality, claim governance, reusable standards, generated registry sources SkillCatalogPort, SkillScorerPort, FactoryAdmissionPort, ClaimEvidencePort, skill/workflow front-door checks Core governance, but sequence after route-critical validation and corpus work. Avoid creating new peer loops.
Runtime and Workspace Adapters BC5 Runtime harness packages, ao CLI driving adapters, git/workspace/session/operator shell, installer and release surfaces HarnessPort, WorkspacePort, OperatorPort, EventBusPort, GitPort, IssueTrackerPort, runtime skill bundles Adapter layer. Keep only surfaces that make the core loop usable across real harnesses.
Orchestration Boundary BC6 Orchestration substrate dispatch, backend selection, agent messaging/wakeups, swarm coordination, Factory driver boundary OrchestrationPort, SwarmDispatchPort, AgentMailPort, ConvergencePort, WorkSpec, SelectionTrace Adapter/substrate boundary. AgentOps owns the loop invariants; the substrate owns when/where/who-supervises.

Bead Alignment

This is a routing snapshot, not a mutation. Verify live status with read-only br before execution.

Disposition Beads Component Reason
KEEP - route-critical ag-tkief, ag-ledger-cache-reconcile-uk3us, ag-verify-close-landed-sweep-33vh1 Work Graph and Loop Engine Ledger truth must be trustworthy before agents can decide what is open, closed, or landed.
KEEP - route-critical ag-wuz9f, ag-o6dya, ag-3l86, ag-8rns, ag-yv25, ag-c3c7 Verification Membrane Red trunk, local release authority, and admission/verdict failures block "verified done."
KEEP - route-critical ag-bl7qf, ag-yk1ff, ag-rcgw4, ag-tkief Work Graph and Loop Engine br/offline tracker migration is not bookkeeping cleanup; it removes fail-open delivery risk.
KEEP - measured moat ag-98g1a, ag-8p8o, ag-epgk, ag-aekxj, ag-2i5jg, ag-k7tq9, ag-lhu0 Context Corpus The corpus claim must become measured retrieval/injection/yield evidence, not file presence.
KEEP - governance ag-3fp54, ag-cuhad, ag-paai, ag-czzf, ag-jc4z, ag-vb3q Skill / Claim Factory New skills/workflows need a governed front door, but this should reduce surfaces rather than expand them.
KEEP AS ADAPTER, then collapse ag-zqtzc, ag-am-read-path-resilience-1anjd, ag-cross-runtime-comms-gateway-sz4c8, ag-am-sibling-deps-build-gate-t1zqq, ag-luklh, ag-am-archive-perf-date-test-6989t Orchestration Boundary / Runtime AM/comms reliability matters only to keep agents coordinated; after the incident, collapse to the one real blocker and close stale siblings.
KEEP AS ADAPTER soc-3csze, ag-s3z, ag-ywk, ag-n90tx, ag-a81zu Runtime and Workspace Adapters Installer/runtime proof matters when it proves real first value; do not let it become another product core.
DEFER ag-v1xk, ag-egpu, ag-egpu.1, ag-dz9xc, ag-xanm, ag-2jp7l, ag-w0wr2 Orchestration Boundary / future substrate Workload objects, reconcile controllers, eGPU, peercred, and meta-orchestration are scale/adversarial capabilities. They do not outrank the route-critical loop.
DEFER or close-candidate ag-o2tc, ag-d6obl, ag-h1l7y, ag-v4wo, ag-w65c, ag-4oj9* Legacy tracker/substrate Dolt/bd hardening is historical unless it is needed to archive or complete the br migration.
TRIM from AgentOps priority ag-i77w, ag-64hr, ag-lvtc, ag-4cfz, ag-4xgl*, ag-real*, ag-yqhs, ag-ecy, ag-3zy, ag-jbre*, ag-b4p*, ag-rtvk, soc-likqx, soc-mtzf, soc-9jaa External domains Relay, business/GTM, family support, career prep, and personal content may be valid work, but they are not AgentOps product components. Track them in their own lane/view.

Trim Rules

Keep work that advances one of these:

  • autonomous code validation and agent-output proof;
  • autonomous goal -> verified done;
  • fail-closed validation and release authority;
  • br-backed work graph truth;
  • measured corpus retrieval/injection/yield;
  • repo-local corpus durability and portability;
  • runtime install/use proof for the core loop.

Defer work when it is real but not needed for the route:

  • hosted or always-on control planes;
  • workload-object/controller APIs;
  • adversarial multi-tenant Linux gates;
  • eGPU/specialized compute;
  • broad AM/ATM/NTM gateways after the active incident is contained;
  • skill polish that lacks a measured failure or route-critical gate.

Trim or move out of the AgentOps product lane when it is:

  • an external product domain such as relay, personal AI, content, GTM, or family support;
  • a duplicate peer loop instead of a step/profile of RPI/evolve;
  • stale bd/Dolt infrastructure after br migration;
  • a generated surface hand-edit;
  • a public claim not backed by current evidence.

Default Sequencing

  1. Reconcile ledger truth and landed/open drift: ag-tkief, ag-ledger-cache-reconcile-uk3us, ag-verify-close-landed-sweep-33vh1.
  2. Restore validation/release health: ag-wuz9f, ag-o6dya, ag-3l86.
  3. Prove the corpus path mechanically: ag-98g1a before ag-8p8o / ag-epgk.
  4. Govern new factory entries: ag-3fp54 / ag-cuhad.
  5. Only then resume broad adapters, substrate, UI, or product-market lanes.

Generated And Curated Surfaces

The generated maps and registries remain derived artifacts. Do not hand-edit registry.json, docs/contracts/context-map.md, docs/reference/agentops-skill-domain-map.md, cli/docs/COMMANDS.md, or docs/cli-surface.*. Change their sources and run the generator/check path instead.

docs/SKILLS.md and skills/SKILL-TIERS.md are curated routers/ledgers, not pure generated projections. Edit their narrative rows deliberately, do not add hard-coded skill counts to docs/SKILLS.md, and let scripts/sync-skill-counts.sh own only the count-bearing markers that remain in skills/SKILL-TIERS.md and other count surfaces.