Review summary

Verdict: Approve the plan docs after rebase — content is strong and aligned with the landed propose (#197). Do not merge as-is without rebasing onto current master.

What actually merges

Against current master, this PR adds only:

• plans/PLAN-HINTS-MCP-CALL-SHAPE.md (+232)
• plans/CURSOR-PROMPTS-HINTS-MCP-CALL-SHAPE.md (+137)

GitHub still shows 66 files / +2058 because the branch was cut before #197–#199 landed (propose, bank-chat fixture, OVERRIDDEN-BY archive). Those commits are already on master — not new work here.

Before merge: rebase onto master so the PR diff and CI reflect the true two-file delta.

PR description

“Planning only — no code, tests, or doc implementation” is accurate vs current master, but the Files-changed tab is misleading until rebase. Worth a one-line note in the PR body.

Plan quality

• Clear problem → solution (pseudo-Python neighbors(['{id}'],…) → MCP-copyable named args; issue #195 combo 1+2+6).
• Good guardrails: no coercion / no ontology bump, HINT_MAX_RENDERED_CHARS = 250, sentinel tests + rg "neighbors\(\['".
• Actionable file matrix and coordination with in-flight DESCRIBE-HINTS-STRUCTURAL-PROPOSE.md.
• Aligns with landed propose/HINTS-MCP-CALL-SHAPE-PROPOSE.md.

Nits (non-blocking)

1. Branch naming: Plan § PR-1 says implementation branch plan/hints-mcp-call-shape — same as this planning branch. Propose suggests cursor/… or feat/…; recommend feat/hints-mcp-call-shape-195 for the implementation PR.
2. Sentinel scope: test_hint_source_no_python_bracket_ids / rg only cover mcp_hints.py + java_ontology.py; plan also edits scripts/generate_edge_navigation.py — consider extending the sentinel or documenting the exclusion.
3. Label consistency: Plan uses PR-1; prompts use PR-HINTS-MCP-CALL-SHAPE-1 — minor.

Next steps

1. Rebase → confirm diff is 2 files → green CI.
2. Merge #201.
3. Implementation: feat/hints-mcp-call-shape-195 off master, follow CURSOR-PROMPTS-HINTS-MCP-CALL-SHAPE.md § PR-HINTS-MCP-CALL-SHAPE-1.