meta: adopt agentic harness framing — update CLAUDE.md, create epics, and plan LAYER-LOG.md

[mdproctor]

Context

The platform strategy has been updated in casehubio/parent (see commits 5692c3b and 9d8e38f). Two significant changes affect all CaseHub apps:

1. All apps are agentic harnesses

casehub-devtown is an agentic harness for software engineering coordination. The domain is PR review and merge orchestration; the structure — coordinating specialist reviewers, enforcing response SLAs, routing adaptively on code content, producing a tamper-evident review record — is the same harness pattern as AML, clinical, and QuarkMind.

Field showcase and tutorial for Java developers in software engineering and DevOps — a domain every Java developer knows from their own daily practice.

Updated docs:

• docs/repos/casehub-devtown.md in parent — now opens with agentic harness description; Tutorial Layers table added (all pending)
• docs/tutorial-strategy.md §7.5 — new section with 7-layer structure and code sketches

2. LAYER-LOG.md is the new definition of done for each layer

As each layer is completed, a structured log entry must be written capturing: what was built, key wiring, gotchas, and a pattern-to-replicate section explicit enough for an LLM to build the equivalent in another domain.

Protocol coming in parent — tracked in casehubio/parent#20.

What to do in this session

• Update CLAUDE.md — add agentic harness framing; reference docs/repos/casehub-devtown.md and tutorial-strategy.md §7.5 for layer structure
• Review open epics — the layered structure (7 layers from tutorial-strategy.md §7.5) should map to child issues. Create the tutorial epic and child issues if they don't exist.
• Plan Layer 1 — naive Java baseline (no CaseHub): direct calls to security analyzer, architecture reviewer, test coverage agent; gap comments showing what's missing
• Create LAYER-LOG.md — empty file ready to receive Layer 1 entry when it ships
• Reference implementation: casehub-aml Layers 1–2 are complete — read the AML code and LAYER-LOG.md before designing devtown Layer 1

Layer structure (from tutorial-strategy.md §7.5)

Layer	Adds	Gap it closes
1	Naive Java	Baseline: direct service calls, no accountability
2	casehub-work	No formal SLA for reviewer response; assignments not tracked
3	casehub-qhorus	No formal obligation per specialist reviewer; DECLINE when outside expertise
4	casehub-ledger	No tamper-evident review record; cannot trace incident to missed finding
5	casehub-engine	Fixed pipeline; no adaptive routing on security flags or architecture changes
6	Trust routing	Experienced reviewers not prioritised on sensitive PRs
7	Comparison vs naive AI code review	—

References

• Parent commits: 5692c3b, 9d8e38f
• docs/repos/casehub-devtown.md in parent
• docs/tutorial-strategy.md §2.0, §2.1b, §7.5 in parent
• AML reference implementation: casehubio/aml (Layers 1–2 complete)