docs: clarify behavior bundle guidance

[Brian Krabach (bkrabach)]

Summary

• Clarify reusable capability behaviors added with --app versus selectable complete roots.
• Preserve the CLI's existing runtime/session behavior while integrating the upstream shared-root lifecycle and truststore dependency updates.
• Keep the goal-batch guidance compact without losing its original triggers, approval planning, verification, mass-change, and continuous-highway boundaries.

Why

This aligns CLI guidance with the merged Foundation #392 prerequisite and the Skills behavior-first model. Portable guidance remains in external bundles; the CLI may compose it but does not own it.

Guidance fidelity

Existing guidance	Final status
Goal-batch independent batch scope and safety/routing	Preserved; the 376-character description retains all five original trigger phrases: “run these in parallel”, “goal-batch”, “launch lanes for these”, “work these N tasks simultaneously”, and “batch these as goals”. It still defines independent goal lanes and bounded safety/routing.
Approval planning	Preserved: plan approval before launch, with the approval boundary retained in the body.
Orchestrator verification	Preserved: the full suite runs after merge; never trust a child to self-claim the guard. The Phase 6 body remains.
Mass-change versus continuous highway	Preserved: per-PR mass-change guidance remains distinct from the continuous-highway workflow and its exclusions.
Prerequisites and mechanisms	Preserved: git, tmux, CLI, goalify, and monitor prerequisites plus the existing scripts/mechanisms remain intact.

The concise description was tightened for catalog usability, not to remove routing semantics.

Intentional scope

• New-host examples use built-in Anchors; reusable capabilities use behavior bundles with --app; selectable roots remain separate.
• Documentation/help guidance changed deliberately. Loader/module implementation, root selection defaults, legacy root body inheritance, provider selection, and tool inheritance were not changed by this guidance correction.
• The integrated upstream also carries shared-root lifecycle/state and truststore changes; those are preserved upstream runtime changes, not claimed as guidance-only changes.

Verification

Gate	Result
Repository validator	validate-bundle-repo full mode 3/3 root/behavior/agent loads; zero severity-ERROR issues; two unchanged skill-description warnings remain: goalify 407 and ten-lane 705
Wheel build	Executed and passed
Integrated CLI tests	85 affected shared-root and guidance tests passed
Installed bundle validation	Directory validation 2/2 clear; agent validation execution succeeded; zero structural ERROR findings, one NO_TOOLS_SECTION warning (quality needs_work because tools are inherited); the valid inherited-tools disposition is preserved rather than adding an empty tools section for cosmetic status
Installed CLI smoke	Version/help and truststore import passed
Foundation runtime	Exact declared pin efd1e26f948bbaef5459d4e5f4321a240ee86b79; public current grader/assets f13d08168e14b5bc4720fbb06c40936eb1a7a7d1

No blanket all-repositories-green claim is made. The earlier Foundation-wide recipe baseline retained unrelated context-size errors outside this three-PR correction set and is not reattributed here.

Review disposition

The independent Fable findings were addressed across the correction set; this is not a formal Fable approval or a blanket model-reliability guarantee. No audit or telemetry machinery was added.

Trade-offs

• Existing hosts can add focused capabilities without replacing their selected root.
• The upstream shared-root lifecycle changes are included for the integrated branch; the guidance surfaces remain explicit about ownership and composition boundaries.

Related PRs

• Merged prerequisite: Foundation #392
• Skills guidance: Skills #77
• Umbrella guidance: amplifier #405

Final correction head

829e3b1afc7926ebee566404fffe091b462fc088.

[Brian Krabach (bkrabach)]

Final correction pass verified at e801080e8298e9f8faaf19a8fe6b096d5d27f6e6: 27 focused tests, structural checks, and Ruff 0.15.7 pass against Foundation 525a730; DTU source/cache parity and the exact installed tree were confirmed. The five independent review findings are addressed. This is a disposition note, not a formal reviewer approval or blanket model-reliability claim.