Files

tegwick bd74d7d122 Document measurement loop plan and ecosystem integration strategy.

Persist INTENT and ecosystem assessments in history/, add ADR-004 for
project metrics with Helix Forge correlation, and register WP-0003 and
WP-0004 workplans with State Hub. Update SCOPE, README, and agency-framework
docs to reflect the two-layer measurement model.

2026-06-16 01:34:13 +02:00

5.4 KiB

Raw Blame History

Ecosystem Integration

How KaizenAgentic composes with adjacent repositories

KaizenAgentic (INTENT.md) defines a meta-improvement layer for coding agents. No single repository implements the full vision. This document describes the two-layer measurement model and integration contracts with ecosystem repos.

Two-Layer Measurement Model

Layer	Question answered	Owner	Storage
Project	How is this agent persona performing in this repo?	kaizen-agentic	`.kaizen/metrics/<agent>/`
Fleet	How are coding sessions performing across repos?	agentic-resources	Helix Forge digest store + baselines

  Coding session (Claude / Codex / Grok)
           │
           ├──────────────────────────────────────┐
           ▼                                      ▼
  agentic-resources                      kaizen-agentic
  (Helix Forge)                          (session close)
  Capture → Digest → Fleet metrics     metrics record → executions.jsonl
           │                                      │
           └──────── helix_session_uid ───────────┘
                         (optional link)

When to use which layer

Project metrics — optimizer recommendations, Coach briefs, per-agent kaizen loop in one codebase (ADR-004).
Fleet metrics — cross-repo friction analysis, pattern distribution, weekly retro, tooling decisions (Helix Forge PRD).

Kaizen-agentic does not re-implement session JSONL ingestion. It may cite Helix session UIDs on project execution records for correlation.

Integration Partners

agentic-resources (P0)

Helix Forge — session capture, fleet aggregates, baselines, weekly retro.

KaizenAgentic	Helix Forge
`.kaizen/metrics/<agent>/executions.jsonl`	Digest store + `measure/baselines.jsonl`
Per-agent persona outcomes	Per-session cross-repo outcomes
`kaizen-agentic metrics optimize`	`session_memory/measure/` aggregates

Correlation fields (ADR-004): helix_session_uid, repo, flavor, tokens, infra_overhead_share.

Workplan: KAIZEN-WP-0004 Part 1.

activity-core (P1)

Event bridge — scheduled and event-driven task creation.

Example ActivityDefinitions (after metrics CLI ships):

Weekly: run kaizen-agentic metrics optimize on repos with .kaizen/
On low success_rate threshold: create review task in issue-core
Post agent install: verify metrics scaffold exists

Workplan: KAIZEN-WP-0004 Part 2.

artifact-store (P1)

Evidence retention — durable registry for generated outputs.

optimizer/analysis.json
recommendations.jsonl snapshots
E2e pilot evidence packages

Retention class: raw-evidence (180d default, aligned with ADR-004).

Workplan: KAIZEN-WP-0004 Part 3.

info-tech-canon (P2)

Semantic canon — agent briefs, patterns, profiles, validation.

Map KaizenAgentTemplate.md → InfoTechCanon profile format
Publish compact agent briefs per persona
Extend kaizen-agentic validate with canon conformance checks

Workplan: KAIZEN-WP-0004 Part 4.

phase-memory (P2, future)

Memory graphs — upgrade from flat memory.md to phased memory profiles.

Fluid memory → project session paths
Stabilized memory → accumulated findings with provenance
Context packages for Coach brief compilation

No WP-0003 blocker; plan after ecosystem integration baseline.

kontextual-engine (P2)

Knowledge operations — ingest wiki/ and agent definitions as governed assets; runtime for KaizenGuidance catalog when built.

llm-connect (P3)

LLM abstraction — use when Coach/optimizer synthesis becomes automated beyond CLI context assembly. Token metrics align with wiki pricing tiers.

domain-tree (P3)

identity-canon (P3)

Terminology for agent persona vs deployed instance vs session actor — supports "digital talent agency" framing without overloading "user".

tele-mcp (TBD)

Listed on Forgejo; not cloned locally. Candidate telemetry MCP adapter for WP-0001 T04. Assess before depending on it.

Boundary Rules

kaizen-agentic owns agent definitions, .kaizen/ conventions, CLI, Coach/optimizer personas, and product framing (INTENT.md, wiki/).
kaizen-agentic does not own session transcript ingestion, task scheduling, artifact bytes, knowledge graph runtime, or LLM providers.
Integrate by contract — ADRs, shared correlation fields, ActivityDefinitions, artifact registration APIs — not by merging repos.
Evidence compounds — fleet baselines inform tooling; project metrics inform agent specs; artifact-store preserves both for audit.

Reading Order

INTENT.md — purpose and boundaries
wiki/EcosystemIntegration.md — this document
docs/adr/ADR-004-project-metrics-convention.md — project metrics schema
history/2026-06-16-ecosystem-assessment.md — full repo comparison
workplans/kaizen-agentic-WP-0004-ecosystem-integration.md — implementation plan

Persisted in history/:

2026-06-16-intent-gap-analysis.md
2026-06-16-ecosystem-assessment.md

5.4 KiB Raw Blame History