docs(skills): slim kernel and curated MCP guidance

[huangzesen]

Summary

• trim duplicated guidance across LingTai kernel manuals and curated MCP skill manuals
• route shared mechanics to real canonical owners while retaining unique safety, authorization, lifecycle, API, credential, notification, scheduler, model, and failure semantics
• keep the change documentation-only and limited to existing Markdown files across the reviewed kernel/MCP skill surface

Scope

• 68 existing Markdown files
• 1,706 insertions / 2,696 deletions
• no add, delete, rename, code, schema, runtime mechanism, config, auth, release, deploy, or cleanup change
• frozen patch SHA-256: d98618d809d4005ca0720d14cd555965b1f60a61f193bfb0db6cd07d0d8a6974

The three blockers found by the first whole-candidate review were repaired narrowly: the email self-send section pointer, the recurring-work route, and three unique Codex CLI facts.

Validation

git diff --check
PASS

python3 -m pytest -q \
  tests/test_validate_skill.py \
  tests/test_skills.py \
  tests/test_catalog_helpers.py \
  tests/test_mcp_skill_manuals.py \
  tests/test_intrinsic_manual_actions.py \
  tests/test_tools_package_data.py \
  tests/test_wheel_sidecar_smoke.py \
  tests/test_docs_governance.py \
  tests/test_preset_runtime_model_docs.py \
  tests/test_environment_variable_catalogue.py \
  tests/test_kernel_update_contract.py \
  tests/test_goal_notification.py \
  tests/test_prompt.py \
  tests/test_nudge_policy.py \
  tests/test_mcp_capability.py \
  tests/test_architecture_documents.py \
  tests/test_read_continuation.py \
  tests/test_vision_capability.py \
  tests/test_knowledge.py \
  tests/test_lingtai_doctor.py \
  tests/test_process_match.py \
  tests/test_notification_tool.py
438 passed in 16.98s

64-directory skill-validator parity
base failing: 7
candidate failing: 7
new failures: 0

Review gates:

• fresh explicit primary-Opus whole-candidate re-review: PASS, P0=0, P1=0
• independent gpt-5.6-terra final whole-candidate review: PASS, P0=0, P1=0
• frozen reviewer snapshot: 68 tracked modifications, zero untracked/ignored input, 68/68 file hashes, exact patch reconstruction

Residual P2 style/discoverability notes remain intentionally out of scope for this one-concern PR.