Second Life MCP v1.0.0

Stable offline MCP tools for LSL and SLua, informed by practical scripting audits
and a complete application migration. No application source or private project
documentation is included in this repository or its distributions.

• Preserve canonical parameter index origins, enum/boolean semantics, permissions
  and purity, with separate metadata provenance. Correct misleading inherited
  SLua indexing advice; link/face enums are not shifted.
• Expand reviewed migration guidance for rounding, UUID validation, vector math,
  list ownership, string/wire formatting, evaluation order, clocks and JSON.
• Add behavioral audit guidance for permission replacement, delayed work,
  request/reply recovery and finite numeric validation. Reference status now
  distinguishes partial caveat coverage and unknown delays from proven safety.
• Exercise generic semantic regressions in the pinned official native SLua VM,
  alongside real MCP transport tests and local compiler/type-checker tests.
• Publish a wheel, source archive and SHA-256 checksums as a stable GitHub release.

The API inventory remains pinned to official definition revision
77cca2b0ee676c46dab7146a2cf63692cc6b91fe. A 1.0 server release does not freeze
SLua or certify viewer/region availability, runtime throttles or application
performance. Compilation never executes submitted MCP client code. Native
runtime probes execute only fixed developer tests and perform no simulator I/O.

Semantic metadata and validation

lookup_symbol now includes parameters[].index_semantics, index_base,
enum_semantics and bool_semantics, plus symbol-level pure,
requires_permissions, return index_base and metadata_source_url / hash.
An unannotated argument has no inferred index origin. Structured semantics come
from the pinned canonical YAML; imported descriptions may retain LSL wording.
Negative indices, not-found sentinels and list-subsequence behavior need their
own migration policy. The server does not automatically rewrite source.

reference_status reports partial behavioral/caveat coverage separately from
official API inventory coverage. Empty caveats are not proof of safety; null
delay means unknown. Practical audits still need queue, permission, memory,
network, asset and lifecycle testing on the intended simulator.

The toolchain fetcher also writes a separate, hash-pinned test-runtime.json
for fixed native developer probes. This file is not accepted as a server
compiler configuration and does not add an MCP execution tool:

second-life-toolchains --output-dir .toolchains
# Set SECOND_LIFE_MCP_TOOLCHAINS to the absolute .toolchains/toolchains.json path.
# Set SECOND_LIFE_MCP_TEST_RUNTIME to the absolute .toolchains/test-runtime.json path.
python -m pytest
second-life-evaluate