Files
pepa-pi-bot/runtime/skills/index.js
T
c910457817 v0.4.0 vNext — closed-loop world model + settlement contract (#29)
* feat(v0.4.0): vNext — closed-loop world model + settlement contract

Implements the vNext architecture from the research doc: demote the noisy
multi-rail planner in favour of a closed loop (world truth → invariant check)
plus a single utility-driven goal authority.

L1 services (fix no_drop / silent pathfinder hang first):
- InventoryLedger: diff-based "did I actually get it" verifier; acquire-food
  now confirms via ledger.gainedSince instead of the unreliable count/event.
- MotionService.gotoSafe: wall-clock timeout + progress watchdog +
  path_update(noPath/timeout) → structured {reached|stuck|timeout|nopath}.

L3 plan — unify the three competing rails (curriculum/manifesto/storyline):
- Settlement Contract: ordered M0–M9 milestones, each invariant-checked
  against an authoritative world view (early steps delegate to the proven
  curriculum; late game adds farming).
- InvariantChecker + predicate library; GoalManager selects the lowest unmet
  milestone via utility argmax (food-urgency preempts, DEPS-style).
- Wired into the scheduler: bot.js precomputes snapshot.contract; reflex.js
  consumes it in place of the storyline rail. Manifesto L0 still preempts.

Eval + robustness:
- Village Score (single 0..1 metric) on the snapshot + TUI "build" line.
- survive.dig-in skill + dusk_dig_in mode (exposed at night, no bed → cover).
- approach_block helper (GoalNear + lookAt, avoids GoalLookAtBlock #341).

+28 new tests (450 total green). LLM remains entirely off the tick path.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* feat(v0.4.0): finish vNext plan — anti-loop, skill-graph, worldDelta diff, flee→motion

Completes the remaining v0.4.0 plan items and one fix motivated by a live
in-game observation (flee hanging 30s against a persistent zombie).

- flee → MotionService.gotoSafe: structured {stuck|timeout|nopath} in ~4s with
  a blind-retreat fallback, instead of the observed 30s pathfinder hang + 3
  watchdog replans. Movements setup guarded so it is unit-testable.
- QW5 anti-loop (runtime/anti-loop.js): same skill failing >=3x in 5min →
  30min blacklist (reflex shouldSkip) + one-shot improvement_request
  (bot.js drainFired -> writeProposal).
- 4.1 closed-loop worldDelta: runSkill snapshots inventory before execute and
  attaches the real delta (_invObserved) to every successful result; opt-in
  skill.expectGain asserts the claimed gain or returns world_unchanged.
- 3.6 skill-graph (Plan4MC): declarative requires/produces for ~20 skills;
  prerequisitesMet/canRun/runnableFrontier; GoalManager annotates suggestions
  with blockedBy when prereqs are unmet.

+22 tests (472 total green). Live smoke confirmed dig-in works and no new
errors; flee loop is what this commit's flee migration addresses.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

---------

Co-authored-by: Yuriy Mayatnikov <mayatnikov@me.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-28 11:29:39 +03:00

340 lines
11 KiB
JavaScript

// Skill substrate. A skill is a small, self-contained, composable unit of
// survival behaviour that the scheduler (today: reflex.js) can call with a
// uniform contract. The contract — required by every skill in this folder:
//
// {
// id: "namespace.action", // stable, machine-readable, e.g. "gather.logs"
// title: "Human label",
// timeoutMs: 45_000, // hard ceiling on execute()
// preconditions(ctx) -> { ok, code?, detail? }
// async execute(ctx, args) -> { ok, code, detail, worldDelta }
// validate?(ctx, result) -> boolean // optional gate after execute
// recover?(ctx, result) -> any | null // optional follow-up hint
// }
//
// The runSkill() wrapper enforces the timeout, normalises the result shape
// (so any caller can rely on the five required fields), runs validate(), and
// calls recover() on failure for the scheduler to consume.
//
// Skills are pure with respect to the runtime — they never read or write
// state-store directly; their `worldDelta` is the only way they communicate
// observed changes back to the scheduler, which then decides what to log.
import { info, warn } from "../log.js";
import { skill as chopLogs } from "./chop-logs.js";
import { skill as eat } from "./eat.js";
import { skill as wander } from "./wander.js";
import { skill as exploreFar } from "./explore-far.js";
import { skill as flee } from "./flee.js";
import { skill as sleep } from "./sleep.js";
import { skill as tunnelOut } from "./recovery-tunnel-out.js";
import { skill as pillarUp } from "./pillar-up.js";
import { skill as digIn } from "./dig-in.js";
import { skill as escapePitSafe } from "./escape-pit-safe.js";
import { skill as diagPhysics } from "./diagnose-physics.js";
import { skill as diagScan, matchSkill as diagMatch } from "./diagnose-scan.js";
import { skill as gatherStone } from "./gather-stone.js";
import { skill as gatherWool } from "./gather-wool.js";
import { skill as acquireFood } from "./acquire-food.js";
import { skill as scoutFood } from "./scout-food.js";
import { skill as relocate } from "./relocate.js";
import { skill as chooseBase } from "./choose-base.js";
import { skill as buildShelter } from "./build-shelter.js";
import { skill as placeChest } from "./place-chest.js";
import { skill as depositSurplus } from "./deposit-surplus.js";
import { skill as farmWheat } from "./farm-wheat.js";
import {
craftPlanksSkill,
craftSticksSkill,
craftWoodenAxeSkill,
craftWoodenPickaxeSkill,
craftWoodenSwordSkill,
craftStoneAxeSkill,
craftStonePickaxeSkill,
craftStoneSwordSkill,
craftFurnaceSkill,
craftChestSkill,
craftTorchSkill,
craftBedSkill,
} from "./craft.js";
const SKILLS = new Map();
function register(skill) {
if (!skill || typeof skill !== "object") throw new Error("skill: not an object");
if (!skill.id || typeof skill.id !== "string") throw new Error("skill: missing id");
if (typeof skill.execute !== "function") throw new Error(`skill ${skill.id}: missing execute`);
if (typeof skill.preconditions !== "function") throw new Error(`skill ${skill.id}: missing preconditions`);
if (SKILLS.has(skill.id)) throw new Error(`skill ${skill.id}: already registered`);
SKILLS.set(skill.id, skill);
}
register(chopLogs);
register(eat);
register(wander);
register(exploreFar);
register(flee);
register(sleep);
register(tunnelOut);
register(pillarUp);
register(digIn);
register(escapePitSafe);
register(diagPhysics);
register(diagScan);
register(diagMatch);
register(gatherStone);
register(gatherWool);
register(acquireFood);
register(scoutFood);
register(relocate);
register(chooseBase);
register(buildShelter);
register(placeChest);
register(depositSurplus);
register(farmWheat);
register(craftPlanksSkill);
register(craftSticksSkill);
register(craftWoodenAxeSkill);
register(craftWoodenPickaxeSkill);
register(craftWoodenSwordSkill);
register(craftStoneAxeSkill);
register(craftStonePickaxeSkill);
register(craftStoneSwordSkill);
register(craftFurnaceSkill);
register(craftChestSkill);
register(craftTorchSkill);
register(craftBedSkill);
export function listSkills() {
return Array.from(SKILLS.values()).map((s) => ({
id: s.id,
title: s.title ?? s.id,
timeoutMs: s.timeoutMs ?? 30_000,
}));
}
export function getSkill(id) {
return SKILLS.get(id) ?? null;
}
// Stable failure codes the wrapper itself can emit. Skills may emit any
// additional codes — but these are the ones runSkill produces.
export const RUNNER_CODES = Object.freeze({
UNKNOWN_SKILL: "unknown_skill",
PRECONDITION_FAILED: "precondition_failed",
TIMEOUT: "timeout",
THREW: "threw",
VALIDATION_FAILED: "validation_failed",
PREEMPTED: "preempted",
DONE: "done",
});
// Signed inventory diff between two count Maps (from InventoryLedger.mark/
// snapshot). Used to attach the real world change to a skill result.
function invDiff(before, after) {
const out = {};
const names = new Set([...(before?.keys?.() ?? []), ...(after?.keys?.() ?? [])]);
for (const n of names) {
const d = (after?.get?.(n) ?? 0) - (before?.get?.(n) ?? 0);
if (d !== 0) out[n] = d;
}
return out;
}
function normaliseResult(res, fallbackCode) {
const ok = !!res?.ok;
return {
ok,
code: res?.code ?? (ok ? RUNNER_CODES.DONE : fallbackCode ?? "failed"),
detail: res?.detail ?? null,
worldDelta: res?.worldDelta ?? null,
};
}
function withTimeout(promise, ms, label) {
let timer;
const timeout = new Promise((_, reject) => {
timer = setTimeout(() => reject(new Error(`${label} timed out after ${ms}ms`)), ms);
});
return Promise.race([promise, timeout]).finally(() => clearTimeout(timer));
}
// v0.3.0-rc.3 — wrap execute() so that if ctx.abortSignal fires we
// stop awaiting (and surface code: "preempted"). The skill itself
// doesn't need to read the signal — the race below ensures runSkill
// returns control to the reflex within one microtask of abort(). The
// skill's own async work may continue in the background harmlessly,
// because the next dispatch will overwrite any shared state.
function raceWithAbort(promise, signal) {
if (!signal) return promise;
if (signal.aborted) {
return Promise.reject(Object.assign(new Error("preempted"), { _preempted: true }));
}
return new Promise((resolve, reject) => {
let settled = false;
const onAbort = () => {
if (settled) return;
settled = true;
reject(Object.assign(new Error("preempted"), { _preempted: true }));
};
signal.addEventListener("abort", onAbort, { once: true });
promise.then(
(v) => {
if (settled) return;
settled = true;
signal.removeEventListener?.("abort", onAbort);
resolve(v);
},
(e) => {
if (settled) return;
settled = true;
signal.removeEventListener?.("abort", onAbort);
reject(e);
},
);
});
}
// Drive one skill through its full lifecycle. The caller (typically reflex.js
// or, eventually, a higher-level scheduler) decides when to invoke; runSkill
// only owns the contract enforcement.
export async function runSkill(id, ctx, args = {}) {
const skill = SKILLS.get(id);
if (!skill) {
warn("skill", `unknown skill ${id}`);
return { ok: false, code: RUNNER_CODES.UNKNOWN_SKILL, detail: id, worldDelta: null };
}
let pre;
try {
pre = skill.preconditions(ctx, args) ?? { ok: true };
} catch (e) {
return {
ok: false,
code: RUNNER_CODES.PRECONDITION_FAILED,
detail: `preconditions threw: ${e.message}`,
worldDelta: null,
};
}
if (!pre.ok) {
const result = {
ok: false,
code: pre.code ?? RUNNER_CODES.PRECONDITION_FAILED,
detail: pre.detail ?? "preconditions failed",
worldDelta: null,
};
if (typeof skill.recover === "function") {
try {
result.recovery = skill.recover(ctx, result) ?? null;
} catch (e) {
warn("skill", `${id}.recover threw: ${e.message}`);
}
}
return result;
}
// WorldDelta diff layer (research §TL;DR): snapshot the inventory before
// execute so we can attach the REAL inventory change to the result and,
// for skills that opt in via `expectGain`, assert the claimed gain actually
// happened instead of trusting the skill's own bookkeeping.
const ledgerBefore = ctx?.ledger?.mark?.() ?? null;
const timeoutMs = skill.timeoutMs ?? 30_000;
let raw;
try {
raw = await withTimeout(
raceWithAbort(skill.execute(ctx, args), ctx?.abortSignal),
timeoutMs,
`skill(${id})`,
);
} catch (e) {
const isTimeout = /timed out after/.test(e.message);
const isPreempted = e?._preempted === true;
const result = {
ok: false,
code: isPreempted
? RUNNER_CODES.PREEMPTED
: isTimeout ? RUNNER_CODES.TIMEOUT : RUNNER_CODES.THREW,
detail: e.message,
worldDelta: null,
};
if (typeof skill.recover === "function") {
try {
result.recovery = skill.recover(ctx, result) ?? null;
} catch (recoverErr) {
warn("skill", `${id}.recover threw: ${recoverErr.message}`);
}
}
return result;
}
const result = normaliseResult(raw);
if (result.ok && typeof skill.validate === "function") {
let valid;
try {
valid = skill.validate(ctx, result);
} catch (e) {
warn("skill", `${id}.validate threw: ${e.message}`);
valid = false;
}
if (!valid) {
const failed = {
ok: false,
code: RUNNER_CODES.VALIDATION_FAILED,
detail: result.detail,
worldDelta: result.worldDelta,
};
if (typeof skill.recover === "function") {
try {
failed.recovery = skill.recover(ctx, failed) ?? null;
} catch (e) {
warn("skill", `${id}.recover threw: ${e.message}`);
}
}
return failed;
}
}
// Closed loop: compare the inventory now vs the pre-execute baseline.
if (result.ok && ledgerBefore && ctx?.ledger) {
try { if (ctx.bot) ctx.ledger.update(ctx.bot); } catch {}
const observed = invDiff(ledgerBefore, ctx.ledger.snapshot());
if (Object.keys(observed).length > 0) {
result.worldDelta = { ...(result.worldDelta ?? {}), _invObserved: observed };
}
// Opt-in strict check: the world must show the claimed gain.
if (skill.expectGain) {
const gain = ctx.ledger.gainedSince(ledgerBefore, skill.expectGain.matcher);
if (gain < (skill.expectGain.min ?? 1)) {
const failed = {
ok: false,
code: "world_unchanged",
detail: `${id} reported ok but ${skill.expectGain.label ?? "expected items"} did not increase (gain ${gain})`,
worldDelta: result.worldDelta,
};
if (typeof skill.recover === "function") {
try { failed.recovery = skill.recover(ctx, failed) ?? null; } catch {}
}
return failed;
}
}
}
if (!result.ok && typeof skill.recover === "function") {
try {
result.recovery = skill.recover(ctx, result) ?? null;
} catch (e) {
warn("skill", `${id}.recover threw: ${e.message}`);
}
}
info("skill", `${id}${result.code}${result.detail ? ` (${JSON.stringify(result.detail).slice(0, 80)})` : ""}`);
return result;
}
// For tests: lets a unit test register a synthetic skill without touching
// the production registry. Returns a teardown function.
export function _registerForTest(skill) {
register(skill);
return () => SKILLS.delete(skill.id);
}