dsh-continual-evolve 0.10.6 → 0.12.0

package/lib/benchmark.js

@@ -1,250 +0,0 @@
-/**
- * Benchmark store: file-backed case definitions and scoreboards under
- * `<baseDir>/evolve/benchmarks/<bid>/`.
- *
- * Layout:
- *   benchmark.json          title, runs (repeats per case), passThreshold
- *   cases/<cid>/statement.md   public task text
- *   cases/<cid>/rubric.json    encrypted scoring criteria (AES-256-GCM, see src/rubric.ts)
- *   scoreboard.json         code-owned aggregates + acceptance history
- *
- * Rubric isolation is code-enforced: rubric plaintext never reaches the
- * disk. Only the evaluation runner decrypts (into the child prompt); the
- * optimizer can read the file and sees ciphertext only. Legacy files that
- * predate encryption carry plaintext and are still readable.
- */
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
-import { join } from "node:path";
-import { encryptRubric, DEV_RUBRIC_KEY, deriveKey } from "./rubric.js";
-/**
- * Close the acceptance loop: when the code-owned decision rejects a
- * candidate refinement, revert it deterministically. The rollback is the
- * same engine path as `/evolve rollback` (inverse edits rebuilt from the
- * applied result — no LLM re-guessing), so it snapshots, versions, and
- * audits like any other mutation. Failures (e.g. the refinement belongs to
- * another session's history) are reported, never thrown: the command shows
- * the manual fallback instead.
- */
-export function rollbackRejectedCandidate(engine, sessionId, candidateId) {
-    try {
-        const result = engine.rollback("local", sessionId, candidateId);
-        const applied = result.appliedEdits.filter((edit) => edit.applied).length;
-        return {
-            rolledBack: true,
-            message: `auto-rollback: reverted refinement ${candidateId} — ${applied} edits restored to the pre-refinement snapshot`,
-        };
-    }
-    catch (cause) {
-        return {
-            rolledBack: false,
-            message: `auto-rollback failed: ${cause instanceof Error ? cause.message : String(cause)} — roll back manually with /evolve rollback <${candidateId}>`,
-        };
-    }
-}
-export function benchmarkDir(baseDir, bid) {
-    return join(baseDir, "evolve", "benchmarks", bid);
-}
-export function sanitizeId(raw) {
-    const id = raw.trim().toLowerCase().replace(/[^a-z0-9]+/g, "_").replace(/^_+|_+$/g, "").slice(0, 40);
-    if (!id) {
-        throw new Error("benchmark id must be a non-empty slug");
-    }
-    return id;
-}
-export function createBenchmark(baseDir, opts) {
-    const id = sanitizeId(opts.title);
-    const dir = benchmarkDir(baseDir, id);
-    if (existsSync(dir)) {
-        throw new Error(`benchmark ${id} already exists`);
-    }
-    const definition = {
-        id,
-        title: opts.title.trim(),
-        description: opts.description?.trim() ?? "",
-        runs: opts.runs ?? 1,
-        passThreshold: opts.passThreshold ?? 60,
-        createdAt: new Date().toISOString(),
-    };
-    mkdirSync(dir, { recursive: true });
-    writeFileSync(join(dir, "benchmark.json"), `${JSON.stringify(definition, null, 2)}\n`, "utf8");
-    writeFileSync(join(dir, "scoreboard.json"), `${JSON.stringify({ candidates: [], decisions: [] }, null, 2)}\n`, "utf8");
-    return definition;
-}
-export function listBenchmarks(baseDir) {
-    const root = join(baseDir, "evolve", "benchmarks");
-    if (!existsSync(root))
-        return [];
-    return readdirSafe(root)
-        .map((id) => loadBenchmark(baseDir, id))
-        .filter((b) => b !== undefined);
-}
-export function loadBenchmark(baseDir, bid) {
-    const path = join(benchmarkDir(baseDir, bid), "benchmark.json");
-    if (!existsSync(path))
-        return undefined;
-    try {
-        return JSON.parse(readFileSync(path, "utf8"));
-    }
-    catch {
-        return undefined;
-    }
-}
-export function addCase(baseDir, bid, title, statement, rubric, rubricKey) {
-    const definition = loadBenchmark(baseDir, bid);
-    if (!definition) {
-        throw new Error(`benchmark ${bid} not found`);
-    }
-    const id = sanitizeId(title);
-    const caseDir = join(benchmarkDir(baseDir, bid), "cases", id);
-    if (existsSync(caseDir)) {
-        throw new Error(`case ${id} already exists in ${bid}`);
-    }
-    mkdirSync(caseDir, { recursive: true });
-    writeFileSync(join(caseDir, "statement.md"), statement, "utf8");
-    // Rubric plaintext never touches the disk; callers pass a resolved key
-    // (config → env → per-installation key file → dev fallback, see
-    // resolveRubricKey) and the dev key here is only a defensive last resort.
-    const stored = rubricKey ? encryptRubric(rubric, rubricKey) : encryptRubric(rubric, deriveKey(DEV_RUBRIC_KEY));
-    writeFileSync(join(caseDir, "rubric.json"), `${JSON.stringify(stored, null, 2)}\n`, "utf8");
-    // A5: initialize case metadata for quality gate.
-    saveCaseMeta(baseDir, bid, id, defaultCaseMeta());
-    return { id, title: title.trim(), statement, rubric };
-}
-export function listCases(baseDir, bid) {
-    const casesDir = join(benchmarkDir(baseDir, bid), "cases");
-    if (!existsSync(casesDir))
-        return [];
-    return readdirSafe(casesDir)
-        .map((id) => {
-        const statementPath = join(casesDir, id, "statement.md");
-        const rubricPath = join(casesDir, id, "rubric.json");
-        if (!existsSync(statementPath) || !existsSync(rubricPath))
-            return undefined;
-        try {
-            const statement = readFileSync(statementPath, "utf8");
-            const rubric = JSON.parse(readFileSync(rubricPath, "utf8"));
-            const meta = loadCaseMeta(baseDir, bid, id);
-            return { id, title: id, statement, rubric, ...(meta ? { status: meta.status } : {}) };
-        }
-        catch {
-            return undefined;
-        }
-    })
-        .filter((c) => c !== undefined);
-}
-export function loadScoreboard(baseDir, bid) {
-    const path = join(benchmarkDir(baseDir, bid), "scoreboard.json");
-    if (!existsSync(path)) {
-        return { candidates: [], decisions: [] };
-    }
-    try {
-        const raw = JSON.parse(readFileSync(path, "utf8"));
-        return {
-            ...(raw.reference ? { reference: raw.reference } : {}),
-            candidates: raw.candidates ?? [],
-            decisions: raw.decisions ?? [],
-        };
-    }
-    catch {
-        return { candidates: [], decisions: [] };
-    }
-}
-export function saveScoreboard(baseDir, bid, board) {
-    writeFileSync(join(benchmarkDir(baseDir, bid), "scoreboard.json"), `${JSON.stringify(board, null, 2)}\n`, "utf8");
-}
-// ── Case meta (gap A5: quality gate + calibration history) ────────────
-export function loadCaseMeta(baseDir, bid, cid) {
-    const path = join(benchmarkDir(baseDir, bid), "cases", cid, "meta.json");
-    if (!existsSync(path))
-        return undefined;
-    try {
-        return JSON.parse(readFileSync(path, "utf8"));
-    }
-    catch {
-        return undefined;
-    }
-}
-export function saveCaseMeta(baseDir, bid, cid, meta) {
-    const metaDir = join(benchmarkDir(baseDir, bid), "cases", cid);
-    mkdirSync(metaDir, { recursive: true });
-    writeFileSync(join(metaDir, "meta.json"), `${JSON.stringify(meta, null, 2)}\n`, "utf8");
-}
-/**
- * Transition a case's lifecycle state. Throws on illegal transitions.
- *   draft → calibrating (start pilot)
- *   calibrating → draft (abandon calibration)
- *   calibrating → frozen (lock baseline)
- */
-export function transitionCaseStatus(baseDir, bid, cid, to) {
-    const meta = loadCaseMeta(baseDir, bid, cid) ?? defaultCaseMeta();
-    const from = meta.status;
-    const valid = (from === "draft" && to === "calibrating") ||
-        (from === "calibrating" && to === "draft") ||
-        (from === "calibrating" && to === "frozen");
-    if (!valid) {
-        throw new Error(`illegal case status transition: ${from} → ${to}`);
-    }
-    meta.status = to;
-    saveCaseMeta(baseDir, bid, cid, meta);
-    return meta;
-}
-function defaultCaseMeta() {
-    return {
-        status: "draft",
-        capability: "",
-        distinguisher: "",
-        shortcuts: "",
-        calibrationHistory: [],
-    };
-}
-/**
- * Quality-check a case: mechanical validation without LLM calls.
- * Returns human-readable problems; empty array means the case passes.
- */
-export function caseCheckProblems(baseDir, bid, cid) {
-    const problems = [];
-    const statementPath = join(benchmarkDir(baseDir, bid), "cases", cid, "statement.md");
-    if (!existsSync(statementPath)) {
-        problems.push("statement.md missing");
-        return problems;
-    }
-    const statement = readFileSync(statementPath, "utf8").trim();
-    if (statement.length < 20) {
-        problems.push(`statement too short (${statement.length} chars, minimum 20)`);
-    }
-    const rubricPath = join(benchmarkDir(baseDir, bid), "cases", cid, "rubric.json");
-    if (!existsSync(rubricPath)) {
-        problems.push("rubric.json missing");
-        return problems;
-    }
-    try {
-        const raw = readFileSync(rubricPath, "utf8");
-        JSON.parse(raw);
-    }
-    catch {
-        problems.push("rubric.json is not valid JSON");
-    }
-    const meta = loadCaseMeta(baseDir, bid, cid);
-    if (!meta) {
-        problems.push("meta.json missing (run /evolve benchmark casecheck to initialize)");
-    }
-    else {
-        if (!meta.capability)
-            problems.push("capability contract is empty");
-        if (!meta.distinguisher)
-            problems.push("distinguisher is empty");
-        if (!meta.shortcuts)
-            problems.push("shortcuts annotation is empty");
-    }
-    return problems;
-}
-import { readdirSync } from "node:fs";
-function readdirSafe(dir) {
-    try {
-        return readdirSync(dir);
-    }
-    catch {
-        return [];
-    }
-}
-//# sourceMappingURL=benchmark.js.map

package/lib/command.d.ts

@@ -1,49 +0,0 @@
-/**
- * The human-facing `/evolve` command: inspect and drive the continual
- * harness from the chat UI without the model in between.
- */
-import type { Context } from "@deepseek-ai/cordis";
-import type { HarnessEntry, HarnessState, RefinementKind } from "./types.js";
-import type { EvolutionEngine } from "./service.js";
-import type { PromotionPolicy } from "./promotion.js";
-import type { RecordLanguagePreference } from "./record-language.js";
-export interface CommandGateOptions {
-    requireGlobalApproval: boolean;
-}
-export interface CommandRuntimeOptions {
-    rubricKey: Buffer;
-    /** When a benchmark decision rejects a candidate, roll the refinement back automatically. */
-    autoRollbackOnReject: boolean;
-    /** P1: capture failed evolution attempts as draft cases in the auto-regression benchmark. */
-    autoCase: boolean;
-    /** Mechanical promotion guards for wrapup/fate (2026-08-22 policy). */
-    promotionPolicy: PromotionPolicy;
-    /**
-     * Initial Memory Agent default from plugin config. The listener is always
-     * registered; this is not a registration gate.
-     */
-    autoReview?: boolean;
-    /** Plugin `recordLanguage` preference for dialog copy (see `approval.ts`). */
-    recordLanguage?: RecordLanguagePreference;
-}
-export declare function registerEvolveCommand(ctx: Context, engine: EvolutionEngine, opts: CommandGateOptions, runtime: CommandRuntimeOptions): void;
-/**
- * Tokenize a command's raw input with shell-like quoting:
- * - a `#` outside quotes starts a comment (rest of the line is dropped);
- * - whitespace separates tokens;
- * - double or single quotes group words into one token and are stripped.
- *
- * This lets users paste help-text examples verbatim, e.g.
- * `/evolve benchmark add-case <bid> "<title>" "<statement>" "<rubric>"`.
- */
-export declare function tokenizeEvolveInput(rawInput: string): string[];
-/** Accept both `<id>` (help-text placeholder form) and bare `id`. */
-export declare function stripAngleBrackets(value: string): string;
-/**
- * Locate an entry by id across every kind of a store. Ids are only unique
- * within a kind, so the lookup scans all four and returns the first match
- * (kind + entry) or undefined. Used by archive/unarchive, which take a bare
- * id from the user.
- */
-export declare function findEntryById(state: HarnessState, id: string): [RefinementKind, HarnessEntry] | undefined;
-//# sourceMappingURL=command.d.ts.map