fix: remove Lua-mod fiction, realign on save-file-injection (ADR-001) model#34
Conversation
… model
Background: 2026-05-08 supervision sprint shipped a 350-line mod.lua and
bespoke scenario.json/Map.json files for Unciv. Subsequent research
(docs/unciv-research-2026-05-11.md) found that Unciv does not support
Lua mod scripting or runtime hooks, and that scenarios are saved-game
files in a scenarios/ folder — not JSON manifests. Both were
architectural fiction that Unciv would not have loaded.
Removed:
- mod/ClaudeKingdoms/mod.lua (350 lines, fictional Lua hook handlers)
- mod/ClaudeKingdoms/scenarios/Tutorial/ (scenario.json + Map.json — wrong shape)
- mod/ClaudeKingdoms/scenarios/LongRunningCampaign/ (same)
- bridge/tests/test_integration.py (Python simulation of the fictional
mod.lua's turn-end hook — the contract it tested doesn't exist)
- 7 cases in bridge/tests/test_tutorial_flow.py that asserted the
deleted scenario JSON files' shape
Renamed / reframed (bridge code that's still sound):
- bridge/tutorial.py: scenario_to_load → scenario_stage. Returns
a stage identifier ("tutorial" / "campaign") instead of a deleted
filename. Old name kept as a back-compat alias.
- bridge/tests/test_per_city_status.py: dropped the Lua-glyph-table
test name in favor of a generic downstream-consumer contract test.
The bridge-side calculation itself is unchanged and correct.
Docs updates:
- mod/ClaudeKingdoms/README.md: documents the actual integration model
(pure JSON content + future external save-edit tool, ADR-001's path).
Adds explicit schema warnings: only Buildings.json and Units.json
have "proper schema" per Unciv docs — others need real-parser
validation.
- CONTRIBUTING.md: contribution path #3 reworded; Lua mention removed
from test ownership matrix; Lua code-style section removed.
- docs/bridge-tui-schema.md: per_city_rewards description corrected —
it's what an external save-edit tool would consume, not what mod.lua
would read at runtime.
- docs/audit-2026-05-08.md: prominent 2026-05-11 retraction header
listing every audit claim that demotes on this finding.
- .claude/handoff.md: full rewrite. Drops "V1 complete" claim,
surfaces the honest acceptance-criteria status (10 ✅ + 5 🟡 + 6 ❌
vs the prior over-stated 18 ✅), and documents the save-edit tool
as the critical-path next deliverable.
Added:
- docs/unciv-research-2026-05-11.md: the source-cited Comet research
that triggered this cleanup.
- .gitignore: logs/ (the daemon writes audit logs there; not for VCS).
Tests: 248 passing (was 262; net -14 from removed fictional tests).
No bridge-side behavior changed.
The Python bridge, TUI, mod JSON content, and onboarding flow are
still real and tested. The bridge↔Unciv integration leg is now
honestly missing rather than dishonestly claimed — it requires an
external save-edit tool against Unciv's actual save format, which is
the project's critical-path next deliverable.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request performs a significant architectural correction by removing fictional Lua-based mod scripting and scenario JSON manifests that are not supported by the Unciv engine. It introduces a research document clarifying that Unciv integration must occur via offline save-file injection and updates the bridge, documentation, and tests to reflect this model. Key feedback includes a correction to a .gitignore rule that inadvertently un-ignored .egg-info directories and a recommendation to update test assertions to use the new scenario_stage API instead of its deprecated alias.
| dist/ | ||
| build/ | ||
| *.egg-info/ No newline at end of file | ||
| *.egg-info/logs/ |
There was a problem hiding this comment.
This change replaces the ignore rule for *.egg-info/ directories with *.egg-info/logs/. This is likely a mistake as it stops ignoring the .egg-info directories themselves, only ignoring a logs subdirectory within them. If the goal was to ignore a top-level logs/ directory, it should be added as a separate line.
*.egg-info/
logs/
| def test_scenario_stage_returns_tutorial_when_not_yet_done(): | ||
| flow = TutorialFlow() | ||
| assert scenario_to_load(flow) == "Tutorial" | ||
| assert scenario_to_load(flow) == "tutorial" |
There was a problem hiding this comment.
The test function has been renamed to test_scenario_stage_... but the assertion still calls the deprecated scenario_to_load alias. To maintain consistency with the new API and ensure the new function is tested directly, please update these calls (lines 159, 161, 168, 174) to use scenario_stage. Note that the import on line 24 will also need to be updated.
1 participant
Summary
2026-05-11 Unciv research (
docs/unciv-research-2026-05-11.md) found that the 2026-05-08 sprint'smod.luaand bespoke scenario JSONs were architectural fiction — Unciv does not support Lua mod scripting or runtime hooks. This PR removes the fiction, reframes the bridge↔Unciv contract on ADR-001's actual save-file-injection model, and honestly demotes the audit claims that depended on the fiction.What's deleted
mod/ClaudeKingdoms/mod.luamod/ClaudeKingdoms/scenarios/Tutorial/+LongRunningCampaign/bridge/tests/test_integration.py(was a Python simulation ofmod.lua)test_tutorial_flow.pyWhat's reframed
bridge/tutorial.py:scenario_to_load→scenario_stage(returns stage id, not a deleted filename)docs/bridge-tui-schema.md: per-city payload is for an external save-edit tool, not mod.luamod/ClaudeKingdoms/README.md: documents the real integration model + schema warningsCONTRIBUTING.md: contribution path feat(spike): implement save/load hook feasibility #3 + test ownership + code style cleaned of Lua refsdocs/audit-2026-05-08.md: 2026-05-11 retraction header listing every claim that demotes.claude/handoff.md: full rewrite — honest acceptance-criteria status (10 ✅ / 5 🟡 / 6 ❌ vs prior over-stated 18 ✅)What's preserved
The Python bridge, TUI, mod JSON content (Nations / Buildings / Beliefs / Tutorials / ModOptions), onboarding flow, frontier helper, audit log, daemon. None of those changed behavior.
Test plan
pytest -q→ 248 passed (was 262; -14 from removed fictional tests, no real behavior changes)grep -rn "mod.lua\|registerHook"returns only the retraction-doc references