Add the 9th modular LLM ability: a turn-transient `TurnResult.disasters`
lever letting the LLM declare natural/man-made disasters (earthquake /
flood / volcanic / wildfire / famine / plague / epidemic) on a nation or
a region, applying a DETERMINISTIC (type, severity) stat dip (no hidden
randomness). Like the hybrid layer it owns no GameState field, needs no
db.ts change, and leaves no durable list — an ongoing plague/epidemic is
simply re-declared each turn it persists.
- shared/protocol.ts: DisasterType / DisasterSeverity / DisasterEvent /
DisastersOps, and `TurnResult.disasters` (between `ideology` and
`markers`; required, default `{ events: [] }`).
- server/abilities/disasters.ts: the ability module (validate / apply /
schemaLine / constraint / isMeaningful / inert stub / empty payload; no
stateFields). `DISASTER_EFFECTS` is the deterministic (type, severity)
dip table (population never touched). Target resolution mirrors hybrid
(a region key hits its SOLE owner; 0/2+ owners dropped) but has NO
`from` field, so there is no self-target drop. Dips clamp to 0-100,
mirror to the target's regions (regionStats), and re-derive the target
nation's stats (same 0-region / no-entry guards).
- Registered last in ABILITIES (after ideology). System-prompt goldens
regenerated (one schema line + one constraint bullet each); the
user.txt golden stays byte-identical (payload() is {}).
- New server/abilities/disasters.test.ts (33 tests); registry.test.ts
updated to the 9-ability contract.
- Secondary cleanups from the v0.40 refactor: AGENTS.md (abilities/ in
the Layout tree + a "how to add a new ability" bullet + a disasters
bullet), delete the orphaned validateTakeovers JSDoc, fold both inline
`as unknown` casts into the top-level ./abilities import, and de-dup
`str` (now imported from ./abilities/shared).
Version bumped 0.40.0 -> 0.41.0.
Behavior-preserving rewire: every per-ability code path now iterates the
ordered ABILITIES registry instead of hand-enumerating the eight fields.
- engine.ts: buildSystemPrompt (schema lines + constraint bullets),
buildUserPrompt (payload keys at their historical positions),
validateTurnResult (per-field validate), applyTurnResult (ordered apply
block), isMeaningful, and the runTurn "unclear" fallback all drive off
ABILITIES. The eight moved ability bodies (~1,249 lines) are deleted;
the names the engine and its tests import are re-exported from
./abilities.
- abilities/types.ts: Ability.stub is a zero-arg method (the stub has no
GameState/AbilityCtx to pass; it needs none) — all eight implementations
were already arg-less.
- stub.ts: deterministicTurn fills the eight ability fields from the
registry (each contributes its inert default), between the core fields
and markers — JSON output stays byte-identical (verified against
origin/master).
- db.ts: the legacy-row migration for ability-owned GameState fields
(wars/trades/embargoes/industry) iterates the registry's stateFields,
only-if-undefined, instead of a hardcoded field list.
No client changes, no new deps, no shared/ runtime deps. typecheck + lint
+ 511 unit tests green; stub JSON byte-identical to origin/master.
Fixes#74
Per-nation named leaders (Nation.leader) and a required TurnResult.leaders
lever ({ changes: [{ target, name, cause }] }). Each change sets the target
nation's leader and applies a deterministic stability dip by cause
(assassination -15, death -3, appointed 0) mirrored to its regions and
re-derived into its stats, plus a narration event (military for an
assassination, political for a death/appointment). An assassination/death
requires the target to already have a leader (a leaderless target is
dropped); an appointment always creates/sets one. Unknown/ambiguous targets
are dropped, never thrown.
- shared/types.ts: optional Nation.leader
- shared/protocol.ts: LeaderCause/LeaderChange/LeaderOps, LEADER_STABILITY_EFFECT,
TurnResult.leaders (required, between nuclear and markers), leader in nationForPrompt
- server/engine.ts: validateLeaderOps, applyLeaderChange, the applyTurnResult
leader block (after nuclear, before markers), isMeaningful, system-prompt
contract + constraint, validateTurnResult + runTurn fallback
- server/stub.ts: leaders always { changes: [] } (e2e unaffected)
- tests: 15 engine.test.ts leader tests + 2 db.test.ts round-trips
- version 0.37.0 -> 0.38.0, CHANGELOG entry
Promote the bare 'war' relation to a named War entity in GameState.wars:
- War: { id, sides[], declaredTurn, declaredDate?, aims?, status, front? }
- TurnResult.wars: { declare: WarDeclareOp[], end: WarEndOp[] }
- syncWarRelations: derive 'war' from active wars, fallback to 'hostile' on end
- Mutual-defense drag-in: belligerent allies added to sides (one level)
- Bare 'war' back-compat: never downgraded, never force-created
- Stub: empty wars (e2e unaffected); ~30 new unit tests
v0.33.0
v0.29.0 — the LLM can now target individual regions' stats via
TurnResult.regionChanges (set/add per stat, clamped 0-100), applied
after nation-wide stateChanges and before the weighted-median
recompute. The map hover tooltip shows each region's own four stats.
The stub stays deterministic (regionChanges: []).
Fixes#46 (PR 2 of 2)
- shared/protocol.ts: TurnResult.unowned (keys released this turn) + the map
presentation contract (regionFillStyle / unownedTooltipLabel, hue-less grey)
- shared/types.ts: required GameState.unowned (cumulative mid-game releases;
[] at init, distinct from preset-unassigned regions)
- server/engine.ts: applyTurnResult applies releases first (sole-owner only,
never fabricating no man's land from unassigned keys), then claims against
the post-release snapshot (gain on an ownerless key = re-claim); lone
release is isMeaningful; prompts + user payload carry the unowned field
- server/stub.ts: unchanged behavior (never releases; e2e stays stable)
- server/db.ts: legacy rows load with unowned defaulted to []
- server/index.ts: create-game sets []; jump reference table includes
released keys so the model can name them for re-claim
- WorldMap/App: unowned prop -> visible grey + 'No man's land' tooltip
- unit tests: 105 -> 127 (release/re-claim suites, presentation contract,
legacy-row normalization)
- llm.ts: LlmClient.chatWithTools — bounded OpenAI function-calling loop
(default 6 tool rounds, then one forced no-tools request); shares the
retry/backoff/reasoning-effort machinery with chat(); tools are only
added to the request body when non-empty, so plain chat() bodies and
their tests are unchanged; a tool-call response is a success (never a
truncation); malformed tool arguments become an error tool-message.
- engine.ts: GET_NEIGHBOURS_TOOL + makeNeighboursExecutor (degrades
unknown tools/args/keys without throwing); playerFrontier (sorted,
deduped, other-nation-owned only); buildSystemPrompt gains a tool
variant (contiguity rules, frontier hint) used only when the tool is
actually sent; buildUserPrompt emits a frontier field whenever a
resolver is given; runTurn takes getNeighbours and uses
chatWithTools when both the resolver and the client's capability are
present, falling back to plain chat() otherwise.
- applyTurnResult: snapshot-first sole-ownership algorithm — a region
key transfers only with exactly one claimant and exactly one previous
owner (different from the gainer); double claims and pre-existing
double ownership are left untouched; lone losses are no-ops; every
key has 0 or 2+ -> 1 owner post-turn, never orphaned or duplicated.
- stub.ts: deterministic one-region contiguous conquest per turn from
the prompt's frontier field (bypasses the tool loop; empty frontier
-> no transfer, all existing assertions stay stable).
- regions.ts: optional region-adjacency.json load; hasAdjacency() and
neighboursOf(keys) (union, deduped, queried excluded, sorted).
- index.ts: jump route passes the adjacency resolver when loaded.
- tests: chatWithTools protocol (echo, executor args, bounded rounds,
malformed args, retry), tool-path runTurn wiring + fallback,
playerFrontier, executor, applyTurnResult invariants, neighboursOf/
hasAdjacency (105 unit tests, 14 e2e — all green).
kamilkrzywda
