Agent workflow: follow the highest-priority requirements in AGENTS.md. Use affected exact checks and searchable Markdown knowledge; ordinary notes do not require full CI or replay.
This project reconstructs the original Japanese TH10.5 version 1.06a executable with reproducible byte comparison as the acceptance criterion.
Note
Reconstruction is still in progress. TH105 is a particularly challenging target, and a faithful, byte-exact reconstruction is expected to take considerable time.
Supply your own resources/th105.exe:
| Property | Required value |
|---|---|
| Version | original Japanese 1.06a |
| Size | 3,129,344 bytes |
| SHA-256 | 56350024879199861579c11b0e1c67b9590e10a8d40cd5996b109deec9afca7e |
| MD5 | 2ae711a6c92c4addbdbf526bc61d8e59 |
| Image base | 0x00400000 |
| Entry point | 0x0068B9D2 |
The official 1.06a updater,
th105_update_106a.exe, contains this executable. The old th105c.exe/1.06
target is different and is deliberately rejected.
scripts/import-target.sh /path/to/th105.exe
scripts/repo-python scripts/verify-target.pyCopyrighted executables and game data are not included.
The project now uses the same conservative control-plane model as the TH08
reconstruction: mapping, source presence, exact matches, and library origin are
separate facts. TH105 keeps a separate provisional boundary/origin ledger
because the initial IDA pass reported 4,001 candidates and VC8 LTCG makes
naïve function/TU accounting unsafe. The current inventory and review count
come from scripts/report-reconstruction-status.py --summary.
The retained source tree came from the former 1.06 effort. Many functions have since been revalidated against 1.06a and accepted into the source and exact ledgers. Unselected retained files remain hypotheses, not 1.06a progress.
Start reconstruction or ledger work with:
scripts/repo-python scripts/verify-target.py
scripts/repo-python scripts/report-reconstruction-status.py --summary
scripts/repo-python scripts/validate-tracking.py --require-targetFor IDA, prefer the Factory-native provider and freshly verify the target, all
five mapped-byte samples and analysis operations as specified in
AGENTS.md and Factory IDA.
scripts/check-ida-mcp.py applies to the legacy bridge, not as a mandatory
prerequisite for an independently attested Factory session. Documentation and
test-workflow maintenance need no target/IDA preflight or native build.
- Current handoff
- Tenshi update resume notes
- Aya root functions
- Reimu root functions
- Local inputs and disposable outputs
- Architecture and TH08/TH105 differences
- Reverse-engineering workflow
- Tool routing
- IDA MCP attestation
- VC8 matching
- Generated progress
- Agent rules
Basic CI needs only Python 3.11+, Git and Bash, with no package installation:
python3 scripts/ci.pyIt checks all tracked Python/JSON/TOML/shell syntax, staged/unstaged whitespace,
ledger consistency, the basic build graph and accidentally tracked private
artifacts. Nonignored untracked files are checked too. GitHub Actions passes
--base COMMIT to check whitespace throughout the PR/push commit range,
including commits behind its merge commit. A clean checkout needs no target or
private tools. Native comparison contracts, compilation, exact comparison and
tests remain task-specific commands.
Use scripts/repo-python for reconstruction and analysis Python commands. It
checks the exact CPython version in .python-version, all package versions in
scripts/requirements-ci.txt, and the importable decoder/emulator native APIs
before executing with that interpreter. Its default environment is .venv;
TH105_PYTHON can select another interpreter meeting the same locks.
Create the isolated environment with uv:
uv venv --python "$(cat .python-version)" .venv
uv pip install --python .venv/bin/python -r scripts/requirements-ci.txtWith the exact configured CPython already installed, python3 -m venv .venv
and .venv/bin/python -m pip install -r scripts/requirements-ci.txt are equivalent
setup steps. GitHub Actions reads the same Python version file but runs only
basic CI, without creating .venv or installing analysis dependencies.
Run scripts/repo-python scripts/progress.py when ledger inputs change.
Progress freshness and historical source pins are not part of basic CI.
The reconstruction loop is a focused VC8 build and whole-function exact
comparison. Allocation-failure/null-pointer coverage, broad execution matrices
and source/checkpoint snapshot tests are not routine acceptance requirements.
Optional analysis/probe tests can be run by filename with
scripts/repo-python -m unittest discover -s tests -p 'test_NAME.py', or together
with scripts/repo-python -m unittest discover -s tests. They diagnose specific
questions and do not establish exactness.
The native connect-menu comparison uses an unmodified Boost 1.40 header closure as a compatible historical reference. On a network-enabled development machine:
scripts/repo-python scripts/upstream_boost.py --fetch
scripts/repo-python scripts/upstream_boost.pyThe source lock pins immutable official commits and every original byte. Fetch is explicit; native builds only verify and fail on missing/altered source. Metadata-only CI does not download or require this SDK. The headers keep their original Boost Software License 1.0. See reference identity and comparison evidence.
The separate original narrow-instance library probe additionally requires
scripts/repo-python scripts/upstream_boost.py --fetch --instances; use
--instances without --fetch for offline verification. See
library origin evidence.
Some original-runtime provenance probes explicitly select Microsoft's KB 971090
archive set. This never changes the active compiler or default link libraries.
Use scripts/repo-python scripts/upstream_vc8_runtime.py to verify it;
--fetch explicitly downloads/extracts the official package without running
the installer. Optional extraction dependencies and pinned source hashes are
documented in runtime provenance.
Repository-authored code and documentation are provided under the MIT License. This does not grant rights to the original game or its assets.
