東方緋想天 ～ Scarlet Weather Rhapsody

This project aims to reconstruct the source code of the original Japanese 東方緋想天 ～ Scarlet Weather Rhapsody version 1.06a executable, with reproducible binary comparison as the acceptance criterion.

The project is in active reverse engineering. The target has been fingerprinted and fully imported into Ghidra; reconstructed routines are counted only after an exact function-byte comparison. The progress graphic above is generated from the machine-readable function ledger.

Target executable

Supply your own original executable as resources/th105.exe:

Property	Required value
Original archive member	th105c.exe
Size	3,039,232 bytes
SHA-256	49c23d9467b9927ba687ed2b873c4bc2d2f39ddadc9f55051ccf10172c0b7c11
PE image base	0x00400000

Localized executables are different binaries and are intentionally rejected. The executable and game data are copyrighted assets and are not included.

scripts/import-target.sh /path/to/th105.rar
python3 scripts/verify-target.py

Reverse-engineering environment

The repository prefers IDA Pro MCP for semantic analysis when a verified IDA session is available, with Ghidra 12.1/GhidraMCP as an independent fallback and current function-inventory authority. Exact acceptance still uses reccmp and objdiff-style object comparison. The workflow borrows the proven mapping/build/report structure of the GensokyoClub TH06 and TH08 projects, while adding a machine-readable function ledger and project-scoped MCP launchers for coding agents.

For IDA-first work, open the exact target, start its MCP plugin, register the stdio bridge as ida-pro-mcp, and run:

python3 scripts/check-ida-mcp.py

If IDA is unavailable, register the th105-ghidra fallback once and verify the complete MCP protocol path with:

scripts/bootstrap-tools.sh
scripts/bootstrap-ghidra-project.sh
scripts/register-codex-mcp.sh
codex mcp list
.tools/src/ghidra-mcp/.venv/bin/python scripts/check-mcp.py

The launcher starts the analyzed headless project on loopback when needed. See docs/MCP.md for setup and security details.

Project map

• Architecture — confirmed binary structure and module plan
• Reverse-engineering workflow — evidence and agent handoff rules
• Workflow evolution decision — gated rollout from strict ledger/comparator foundations through packets, clone fan-out, objdiff, linking, and runtime validation
• IDA-first analysis — exact routing, boundary safety, and Ghidra fallback
• Reconstruction map — unlock-first gameplay and character tree
• Gameplay reconstruction framework — core lanes, ABI/type contracts, dependency graph, and agent-ready worklist
• Matching workflow — compiler, reccmp, and objdiff stages
• Exact-matching pattern catalog — reusable VC8, COFF, relocation, and source-shaping techniques
• Exact-matching skill — focused mismatch classification and verified UI/input-era VC8 case studies
• UI reconstruction framework — scenario-select ABI, call graph, and shared menu helpers
• Progress — generated from config/functions.csv
• AGENTS.md — mandatory operating rules for coding agents
• Parallel-agent skill — safe, high-throughput IDA/Ghidra/ledger delegation

Regenerate and validate the tracking data with:

python3 scripts/generate-tracking.py
python3 scripts/core-worklist.py --check
python3 scripts/progress.py --check
python3 scripts/validate-tracking.py

Build status

The PE Rich header and linker metadata point to Visual C++ 2005 (VC8), including 42 C++ LTCG records. VC8 SP1 already reproduces the tracked matching routines exactly, but the original service level, complete flags, translation-unit partition, and link order are still being established. A matching full executable is not yet advertised. See docs/BUILD_MATCHING.md.

License

Repository-authored code and documentation are provided under the MIT License. This does not grant rights to the original game or its assets.