[ThatRendle]

Summary

Stops the test suite writing sessions into the developer's real ~/.dmon/sessions, and settles what marks a dmon project root.

• Root marker settled: a project root is a directory containing .dmon/config.yaml, the rule SessionDirectoryResolver already implements. The alternative, a bare .dmon/ directory, was rejected: ActiveModelStore writes config.local.yaml on a model switch, which would quietly relocate sessions.
• Spec + ADR: the session-storage discovery requirement is replaced. It uses .dmon naming instead of the stale .daemon, the config-file marker, and the bootstrap the code actually performs (~/.dmon/, not CWD). It states only what is true and decided. A root's sessionStore is ignored from a subdirectory; that case is deliberately unspecified and parked in tech-debt/session-store-setting-ignored-from-subdirectory.md. ADR-004 gets an amendment note only.
• Three writers fixed. Each one's config.yaml marker now explicitly pins sessionStore: local:
  • LiveToolCallE2ETest: a session on every live run. It now asserts the session lands in its temp root, and it is tagged Category=Live, so plain make test no longer makes a paid call.
  • CoreProcessFixture / IntegrationSmokeTest: the empty "twin" session on every make test. It now asserts the session lands in the fixture's root (this runs in CI).
  • CoreProcessManagerRestartTests: writes to ~ only when the global config sets sessionStore: global or a path, so a store diff couldn't see it (design D9).
• Rule pinned by SessionDirectoryResolverTests: a .dmon/ holding only config.local.yaml is not a root (mutation-checked).

No production code changes. No user sessions move.

Verification

Whole-suite check (task 3.2): no dmon host process and nothing on port 8666. Both session stores were counted with find -type d, before and after.

Pass	non-Live	Live	home store	repo store	canary
1, real stores	exit 0, every assembly complete (Core 627 passed / 1 skipped, Terminal 194/194)	exit 0; LiveToolCallE2ETest Passed (not Skipped) with a real Gemini call	856 → 856	770 → 770	n/a
2, canary sessionStore=<scratch>/canary-store	exit 0, same counts	exit 0	856 → 856	770 → 770	never created

The canary proves the point only because ~/.dmon/config.yaml sets no sessionStore; I confirmed that (the environment is the lowest configuration layer). openspec validate session-root-resolution --strict passes. All three sections were approved by the section supervisor.

Found along the way (registered, not fixed here)

• tech-debt/terminal-tests-hang.md, cause verified: the three InitCommandTests hang because InitFeedFixture awaits pack-core.sh stdout EOF with no timeout, while a reusable MSBuild node spawned by dotnet pack holds the pipe open until it idles out (~15 min). With MSBUILDDISABLENODEREUSE=1 they pass in 7 s; the control hangs. Core's ComposedCoreFeedFixture is a line-for-line twin and the leading suspect for the intermittent Core hang. Workaround until fixed: export MSBUILDDISABLENODEREUSE=1 for suite runs.
• New notes: meko-live-smoke-passes-without-key.md (make test-live reports Passed having tested nothing) and repo-dmon-config-unpinned.md (a latent hazard, tied to the subdirectory question).

After merge

• Amend tech-debt/live-e2e-test-writes-into-home-session-store.md's status line with the merge commit.
• /opsx:archive session-root-resolution.
• Pruning the test sessions that have already piled up in ~/.dmon/sessions is left to the owner.

Change: session-root-resolution

🤖 Generated with Claude Code

https://claude.ai/code/session_015CJu5CxqY5B6o3GfMVAfsu