chore: audit orchestrator conformance with skill-forward architecture

## Summary

Audit current user-facing orchestrators against the skill-forward artifact architecture established by HVE Builder and RPI.

The target allocation is:

* Skills own reusable workflows, domain knowledge, references, templates, and scripts.
* Agents provide user-selected multi-turn or bounded autonomous orchestration and remain thin over reusable capabilities.
* Subagents own isolated, high-volume, parallel, fresh-context, model-specific, or verification work and remain background-only unless independent user value is explicitly demonstrated.
* Instructions carry durable path-scoped conventions rather than on-demand workflow capability.

This issue supersedes #1310. The dual-mode Code Review pattern that motivated that issue was retired by #2100, and #2474 established the current reusable-skill, thin-wrapper, and bounded-worker architecture.

## Children

* #1311: define skill-forward orchestrator conformance guidance
* #1312: audit orchestrators for skill-forward conformance

## Scope

1. Clarify the skill-forward orchestrator decision rules in current HVE Builder guidance.
2. Inventory current collection-distributed user-facing agents that coordinate skills, instructions, or subagents.
3. Audit applicable orchestrators for responsibility placement, capability duplication, wrapper thinness, worker isolation, and worker invocability.
4. Record an evidence-backed disposition for every audited system.
5. Create focused follow-up issues only for actionable findings not already tracked.

Dual-mode agent conversion is not a goal. A user-invocable worker is an exception requiring independently demonstrated user value, not the default answer to reusable capability.

## Reference Implementations

Use the current RPI, Documentation, and Code Review architectures as comparison points without assuming every orchestrator must use an identical shape.

## Acceptance Criteria

* [ ] Current HVE Builder guidance defines the responsibility boundary among skills, user-facing agents, instructions, and background-only subagents.
* [ ] The guidance defines what a thin wrapper may legitimately own, including conversation state, phase sequencing, gates, handoffs, capability bindings, and user confirmations.
* [ ] The guidance states that reusable workflow and domain capability should not be duplicated in agent wrappers.
* [ ] The guidance states that background-only workers use `user-invocable: false` and defines the evidence required for an exception.
* [ ] The audit inventories all current collection-distributed user-facing orchestration agents and identifies which warrant detailed assessment.
* [ ] The detailed audit includes, at minimum, RPI, Code Review, Security Reviewer, DT Coach, ADO/GitHub/Jira backlog management, RAI Planner, and SSSC Planner.
* [ ] Each audited system receives one disposition: conformant/no change, guidance-only correction, extract reusable capability to a skill, thin the user-facing wrapper, correct background-worker exposure, or justified exception.
* [ ] Every disposition cites current file evidence and records a no-change rationale where no refactoring is recommended.
* [ ] Remediation work is captured in deduplicated follow-up issues rather than implemented implicitly as part of the audit.
* [ ] Rescoped issues #1311 and #1312 are completed and linked to this parent.

## Related

* Supersedes #1310
* Historical origin: #1304
* Code Review consolidation: #2100
* Skill-forward RPI consolidation: #2474