* feat(mcp): add compact input schema profile for size-capped providers Moonshot's function-calling API rejects a request while any single tool parameters schema exceeds its literal serialized size cap, so clients such as kimi-cli fail every request while the default catalog is attached (#1484). Add an opt-in REA_MCP_INPUT_SCHEMA_PROFILE=compact server presentation. Each advertised input schema is rendered within a 13,312-byte budget, measured to sit below the observed provider cap: annotation prose (nested descriptions, examples, defaults, titles) is dropped, repeated annotation-free subschemas are shared through schema-local $defs/$ref references the provider validates without expansion, and the three structurally oversized schemas are advertised in a documented reduced form. The canonical catalog, tool set, output schemas, and server-side canonical validation are unchanged; the default full profile is byte-identical to today's advertisement. Boundary tests pin every compact schema to the budget, keep them valid JSON Schema that still accepts every canonical example, hold reduced presentations at three or fewer, and verify canonical rejection still guards a permissive advertisement. Verified against the real Moonshot API through kimi-cli: prompts succeed with the compact catalog attached where the default catalog fails every request with 400. * fix(mcp): preserve compact schemas through Kimi conversion --------- Co-authored-by: morluto <76467478+morluto@users.noreply.github.com> |
||
|---|---|---|
| .. | ||
| android | ||
| c | ||
| contracts | ||
| ghidra | ||
| ida | ||
| inspector | ||
| interface-builder | ||
| napi | ||
| native | ||
| objc | ||
| providers | ||
| readiness | ||
| swift | ||
| versions | ||
| README.md | ||
| readmeConsistency.test.ts | ||
| releaseConfiguration.test.ts | ||
| scanTodos.test.ts | ||
README.md
Source-built conformance fixtures
These sources are deliberately small semantic oracles for REA providers. Run
npm run verify:fixtures on macOS or Linux to compile them into the ignored
build/conformance/ directory, emit a hash-and-toolchain manifest, and verify
their platform-native symbols and strings with local command-line tools. The
portable C, version, and large-inventory fixtures build as Mach-O on macOS and
ELF on Linux; Objective-C, Swift, and N-API-like fixtures remain macOS-only.
Only source files are versioned. Generated binaries, generated large-fixture
source, debug information, and manifests must remain under build/.
ghidra/inventory.c is also compiled into temporary Linux ELF debug and
stripped variants by npm run verify:ghidra. It fixes local functions, external
puts linkage, defined strings, and writable/executable memory needed to prove
the read-only Ghidra inventory contracts without committing a binary.
scripts/create-ghidra-windows-fixture.mjs separately emits a deterministic,
ignored native x86-64 PE application for the controlled Windows Ghidra P0 lane.
The generator and fixed SHA-256 are versioned; the .exe is never committed or
executed. npm run verify:ghidra:windows uses it to prove all 19 operations,
digest linkage, transport, and cleanup on the self-hosted real-Ghidra runner.
The C fixture fixes a known call chain and strings. The version pair fixes an added symbol, changed string, and changed call relationship. The generated large fixture contains 1,205 uniquely named functions and strings to verify that providers return complete inventories. Objective-C, Swift, and N-API-like fixtures provide metadata-oriented inputs without package or network dependencies.
readiness/ adds source-only generated JavaScript CLI and Electron fixtures for
the repository's reconstruction-readiness verifier. npm run verify:readiness
combines them with the compiled native fixtures, exercises CLI and MCP parity
for real analysis operations, and checks the internal machine-readable report
documented in docs/reconstruction-readiness.md.