[LE0-Lin]

Cursor project rules can attach based on the files in an agent's context. A repository-wide text comparison cannot explain which rules a particular task would trigger, and disjoint globs can still co-activate in a multi-file task.

Add the opt-in, unscored scope PATH --target FILE command and cursor-file-context-v1 model. It parses a bounded scalar metadata grammar, matches declared logical paths using an iterative glob engine, retains manual/model/unsupported cases as unknown, and reports textual opposing-directive signals with original source lines and trigger evidence. Raw metadata/directive bodies are omitted from reports; the new runtime module is included in tool fingerprints.

Includes a runnable synthetic example, installed-wheel journeys for both aliases and the module entrypoint, a detailed grammar/reference, and synchronized English/Simplified/Traditional core guides. Existing scan/diff/compare findings and gate scores are unchanged. This first phase does not observe Cursor, prove semantic conflicts, implement effective-policy inheritance/change-impact analysis, or update published v0.24.0 artifacts/Action refs. Whole-repository supported-source preflight still applies.

Validation:

• Windows: 572 tests, 560 passed and 12 platform skips.
• WSL Linux: all 28 new model/integration tests passed.
• Built wheel + sdist; wheel-only installed acceptance passed all 17 journeys, including three-entrypoint scope behavior, multi-file co-activation, unknown metadata, source lines, privacy and error handling.
• Existing demo remains F 56; repository self-scan remains A 100; staged diff whitespace check passed.
• Manual pinned-repository smoke on c334dd38fd6dcf06ac7d2d6374a2abaa94f89b31: five completed scans and one expected typed rejection for the pinned Reaparr dangling source; no acquisition errors. All maintained expectations match. The retained report identity matches this 17-module runtime (af0ca24b63f7d502711e0addbab8989ec6f839ea74a2af54d3e681cb4d84a5f7). This is compatibility evidence, not six successful scans or an independent accuracy estimate.

These are deterministic development and installation checks, not independent accuracy or user-adoption evidence.