@zhixuan92/multi-model-agent-core 4.7.5 → 4.7.7

package/dist/reporting/verify-stage-runner.js

@@ -1,102 +0,0 @@
-import { spawn } from 'node:child_process';
-const DEFAULT_TIMEOUT_MS = 300_000; // 5 min
-const MAX_OUTPUT_BYTES = 64 * 1024; // 64 KB
-class RollingTail {
-    chunks = [];
-    bytes = 0;
-    push(s) {
-        const buf = Buffer.from(s, 'utf8');
-        this.chunks.push(buf);
-        this.bytes += buf.length;
-        while (this.chunks.length > 0 && this.bytes - this.chunks[0].length >= MAX_OUTPUT_BYTES) {
-            this.bytes -= this.chunks.shift().length;
-        }
-    }
-    toString() {
-        if (this.bytes === 0)
-            return '';
-        const concat = Buffer.concat(this.chunks);
-        if (concat.length <= MAX_OUTPUT_BYTES)
-            return concat.toString('utf8');
-        return concat.subarray(concat.length - MAX_OUTPUT_BYTES).toString('utf8');
-    }
-}
-export class VerifyStageRunner {
-    /**
-     * Run a post-implementation verify command (e.g. `npm test`).
-     *
-     * IMPORTANT: `command` is executed via `shell: true` and MUST be trusted input
-     * (config-supplied, not user-supplied). Do not pass unsanitized user input here.
-     */
-    async run(command, cwd, options = {}) {
-        const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
-        const start = Date.now();
-        return new Promise((resolve, reject) => {
-            const proc = spawn(command, { shell: true, cwd });
-            const stdout = new RollingTail();
-            const stderr = new RollingTail();
-            let resolved = false;
-            const finish = (result) => {
-                if (!resolved) {
-                    resolved = true;
-                    resolve(result);
-                }
-            };
-            const timer = setTimeout(() => {
-                finish({
-                    exitCode: -1,
-                    stdout: stdout.toString(),
-                    stderr: stderr.toString(),
-                    signal: null,
-                    durationMs: Date.now() - start,
-                    errorCode: 'validator_verify_command_failed',
-                });
-                try {
-                    proc.kill('SIGKILL');
-                }
-                catch {
-                    // best-effort kill
-                }
-            }, Math.max(1, timeoutMs));
-            proc.stdout?.on('data', (d) => stdout.push(d.toString()));
-            proc.stderr?.on('data', (d) => stderr.push(d.toString()));
-            proc.on('error', (err) => {
-                clearTimeout(timer);
-                reject(err);
-            });
-            proc.on('exit', (code, sig) => {
-                clearTimeout(timer);
-                if (sig) {
-                    finish({
-                        exitCode: code ?? -1,
-                        stdout: stdout.toString(),
-                        stderr: stderr.toString(),
-                        signal: sig,
-                        durationMs: Date.now() - start,
-                        errorCode: 'validator_verify_command_failed',
-                    });
-                }
-                else if (code === 0) {
-                    finish({
-                        exitCode: 0,
-                        stdout: stdout.toString(),
-                        stderr: stderr.toString(),
-                        signal: null,
-                        durationMs: Date.now() - start,
-                    });
-                }
-                else {
-                    finish({
-                        exitCode: code ?? -1,
-                        stdout: stdout.toString(),
-                        stderr: stderr.toString(),
-                        signal: null,
-                        durationMs: Date.now() - start,
-                        errorCode: 'validator_verify_command_failed',
-                    });
-                }
-            });
-        });
-    }
-}
-//# sourceMappingURL=verify-stage-runner.js.map

package/dist/reporting/verify-stage-runner.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"verify-stage-runner.js","sourceRoot":"","sources":["../../src/reporting/verify-stage-runner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAE3C,MAAM,kBAAkB,GAAG,OAAO,CAAC,CAAC,QAAQ;AAC5C,MAAM,gBAAgB,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,QAAQ;AAe5C,MAAM,WAAW;IACP,MAAM,GAAa,EAAE,CAAC;IACtB,KAAK,GAAG,CAAC,CAAC;IAElB,IAAI,CAAC,CAAS;QACZ,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;QACnC,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACtB,IAAI,CAAC,KAAK,IAAI,GAAG,CAAC,MAAM,CAAC;QACzB,OAAO,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,IAAI,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,IAAI,gBAAgB,EAAE,CAAC;YACxF,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,MAAM,CAAC,KAAK,EAAG,CAAC,MAAM,CAAC;QAC5C,CAAC;IACH,CAAC;IAED,QAAQ;QACN,IAAI,IAAI,CAAC,KAAK,KAAK,CAAC;YAAE,OAAO,EAAE,CAAC;QAChC,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAC1C,IAAI,MAAM,CAAC,MAAM,IAAI,gBAAgB;YAAE,OAAO,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QACtE,OAAO,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,GAAG,gBAAgB,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IAC5E,CAAC;CACF;AAED,MAAM,OAAO,iBAAiB;IAC5B;;;;;OAKG;IACH,KAAK,CAAC,GAAG,CAAC,OAAe,EAAE,GAAW,EAAE,UAAoC,EAAE;QAC5E,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,kBAAkB,CAAC;QAC1D,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAEzB,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;YACrC,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,CAAC,CAAC;YAClD,MAAM,MAAM,GAAG,IAAI,WAAW,EAAE,CAAC;YACjC,MAAM,MAAM,GAAG,IAAI,WAAW,EAAE,CAAC;YACjC,IAAI,QAAQ,GAAG,KAAK,CAAC;YAErB,MAAM,MAAM,GAAG,CAAC,MAAoB,EAAQ,EAAE;gBAC5C,IAAI,CAAC,QAAQ,EAAE,CAAC;oBACd,QAAQ,GAAG,IAAI,CAAC;oBAChB,OAAO,CAAC,MAAM,CAAC,CAAC;gBAClB,CAAC;YACH,CAAC,CAAC;YAEF,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;gBAC5B,MAAM,CAAC;oBACL,QAAQ,EAAE,CAAC,CAAC;oBACZ,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE;oBACzB,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE;oBACzB,MAAM,EAAE,IAAI;oBACZ,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;oBAC9B,SAAS,EAAE,iCAAiC;iBAC7C,CAAC,CAAC;gBACH,IAAI,CAAC;oBACH,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBACvB,CAAC;gBAAC,MAAM,CAAC;oBACP,mBAAmB;gBACrB,CAAC;YACH,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC,CAAC;YAE3B,IAAI,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,CAAkB,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;YAC3E,IAAI,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,CAAkB,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;YAE3E,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,EAAE,EAAE;gBACvB,YAAY,CAAC,KAAK,CAAC,CAAC;gBACpB,MAAM,CAAC,GAAG,CAAC,CAAC;YACd,CAAC,CAAC,CAAC;YAEH,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,GAAG,EAAE,EAAE;gBAC5B,YAAY,CAAC,KAAK,CAAC,CAAC;gBACpB,IAAI,GAAG,EAAE,CAAC;oBACR,MAAM,CAAC;wBACL,QAAQ,EAAE,IAAI,IAAI,CAAC,CAAC;wBACpB,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE;wBACzB,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE;wBACzB,MAAM,EAAE,GAAG;wBACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;wBAC9B,SAAS,EAAE,iCAAiC;qBAC7C,CAAC,CAAC;gBACL,CAAC;qBAAM,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;oBACtB,MAAM,CAAC;wBACL,QAAQ,EAAE,CAAC;wBACX,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE;wBACzB,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE;wBACzB,MAAM,EAAE,IAAI;wBACZ,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;qBAC/B,CAAC,CAAC;gBACL,CAAC;qBAAM,CAAC;oBACN,MAAM,CAAC;wBACL,QAAQ,EAAE,IAAI,IAAI,CAAC,CAAC;wBACpB,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE;wBACzB,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE;wBACzB,MAAM,EAAE,IAAI;wBACZ,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;wBAC9B,SAAS,EAAE,iCAAiC;qBAC7C,CAAC,CAAC;gBACL,CAAC;YACH,CAAC,CAAC,CAAC;QACL,CAAC,CAAC,CAAC;IACL,CAAC;CACF"}

package/dist/review/run-reviewer.d.ts

@@ -1,40 +0,0 @@
-import type { ExecutionContext } from '../lifecycle/lifecycle-context.js';
-import type { AgentType } from '../types.js';
-export type RunReviewerInput = {
-    prompt: string;
-    ctx: ExecutionContext;
-    reviewer: 'spec' | 'quality';
-    /**
-     * The implementer's tier. The reviewer runs on the OPPOSITE tier as a
-     * "second-opinion needs a different perspective" policy:
-     *   implementer=standard → reviewer=complex (capable reviewer of cheap work)
-     *   implementer=complex  → reviewer=standard (cheap sanity check of expensive work)
-     * If the inverted tier has no provider configured, falls back to the
-     * implementer tier and records a `validation_warnings` diagnostic upstream.
-     */
-    implementerTier: AgentType;
-};
-/**
- * Cross-tier inversion: reviewer tier is the opposite of the implementer's.
- * Exported so callers can compute the tier independently (e.g. for logging
- * or for the `mergeStageStats` call in review-handler.ts).
- */
-export declare function invertedReviewerTier(implementerTier: AgentType): AgentType;
-export type RunReviewerResult = {
-    kind: 'ok';
-    text: string;
-    costUSD: number | null;
-    turnsUsed: number;
-    ms: number;
-    model: string | null;
-    inputTokens: number;
-    outputTokens: number;
-    cachedReadTokens: number;
-    cachedNonReadTokens: number;
-} | {
-    kind: 'transport_error';
-    message: string;
-    ms: number;
-};
-export declare function runReviewerTurn(input: RunReviewerInput): Promise<RunReviewerResult>;
-//# sourceMappingURL=run-reviewer.d.ts.map

package/dist/review/run-reviewer.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"run-reviewer.d.ts","sourceRoot":"","sources":["../../src/review/run-reviewer.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mCAAmC,CAAC;AAC1E,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAG7C,MAAM,MAAM,gBAAgB,GAAG;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,gBAAgB,CAAC;IACtB,QAAQ,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B;;;;;;;OAOG;IACH,eAAe,EAAE,SAAS,CAAC;CAC5B,CAAC;AAEF;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,eAAe,EAAE,SAAS,GAAG,SAAS,CAE1E;AAED,MAAM,MAAM,iBAAiB,GACzB;IACE,IAAI,EAAE,IAAI,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,mBAAmB,EAAE,MAAM,CAAC;CAC7B,GACD;IAAE,IAAI,EAAE,iBAAiB,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,CAAC;AAE7D,wBAAsB,eAAe,CAAC,KAAK,EAAE,gBAAgB,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAqCzF"}

package/dist/review/run-reviewer.js

@@ -1,55 +0,0 @@
-// Thin wrapper that runs one reviewer-tier LLM turn with 3-attempt transport-retry policy.
-// Spec §4.6: transport failures (network, 5xx, model gateway, timeout) → 3 total attempts
-// with backoff 0s → 1s → 2s. Non-transport caps (turn_cap, sandbox) → no retry.
-import { HUMAN_LABEL } from '../lifecycle/stage-labels.js';
-/**
- * Cross-tier inversion: reviewer tier is the opposite of the implementer's.
- * Exported so callers can compute the tier independently (e.g. for logging
- * or for the `mergeStageStats` call in review-handler.ts).
- */
-export function invertedReviewerTier(implementerTier) {
-    return implementerTier === 'complex' ? 'standard' : 'complex';
-}
-export async function runReviewerTurn(input) {
-    const t0 = Date.now();
-    const backoffMs = [0, 1000, 2000]; // attempt 1: no backoff; 2: 1s; 3: 2s
-    let lastErr = '';
-    for (let attempt = 0; attempt < 3; attempt++) {
-        if (backoffMs[attempt] > 0)
-            await sleep(backoffMs[attempt]);
-        try {
-            // Cross-tier inversion. If the inverted tier has no provider
-            // configured (single-tier deployments), fall back to the implementer
-            // tier so the reviewer can still run. The fallback is observable
-            // via the resolved `session.model` carried into the wire payload.
-            const desired = invertedReviewerTier(input.implementerTier);
-            const providers = input.ctx.providers;
-            const tierToUse = providers && providers[desired] ? desired : input.implementerTier;
-            const session = input.ctx.getSession(tierToUse);
-            const r = await session.send(input.prompt, { stageLabel: HUMAN_LABEL.review });
-            return {
-                kind: 'ok',
-                text: r.output ?? '',
-                costUSD: r.costUSD ?? null,
-                turnsUsed: r.turns ?? 1,
-                ms: Date.now() - t0,
-                model: typeof r.model === 'string'
-                    ? r.model
-                    : null,
-                inputTokens: r.usage?.inputTokens ?? 0,
-                outputTokens: r.usage?.outputTokens ?? 0,
-                cachedReadTokens: r.usage?.cachedReadTokens ?? 0,
-                cachedNonReadTokens: r.usage?.cachedNonReadTokens ?? 0,
-            };
-        }
-        catch (err) {
-            lastErr = err instanceof Error ? err.message : String(err);
-            // Only retry on transport-class errors; turn_cap / sandbox are non-retryable.
-            if (!/transport|network|5\d\d|timeout/i.test(lastErr))
-                break;
-        }
-    }
-    return { kind: 'transport_error', message: lastErr || 'reviewer failed', ms: Date.now() - t0 };
-}
-function sleep(ms) { return new Promise(r => setTimeout(r, ms)); }
-//# sourceMappingURL=run-reviewer.js.map

package/dist/review/run-reviewer.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"run-reviewer.js","sourceRoot":"","sources":["../../src/review/run-reviewer.ts"],"names":[],"mappings":"AAAA,2FAA2F;AAC3F,0FAA0F;AAC1F,gFAAgF;AAIhF,OAAO,EAAE,WAAW,EAAE,MAAM,8BAA8B,CAAC;AAiB3D;;;;GAIG;AACH,MAAM,UAAU,oBAAoB,CAAC,eAA0B;IAC7D,OAAO,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;AAChE,CAAC;AAiBD,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,KAAuB;IAC3D,MAAM,EAAE,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACtB,MAAM,SAAS,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC,sCAAsC;IACzE,IAAI,OAAO,GAAG,EAAE,CAAC;IACjB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,OAAO,EAAE,EAAE,CAAC;QAC7C,IAAI,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC;YAAE,MAAM,KAAK,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;QAC5D,IAAI,CAAC;YACH,6DAA6D;YAC7D,qEAAqE;YACrE,iEAAiE;YACjE,kEAAkE;YAClE,MAAM,OAAO,GAAG,oBAAoB,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC;YAC5D,MAAM,SAAS,GAAI,KAAK,CAAC,GAA2D,CAAC,SAAS,CAAC;YAC/F,MAAM,SAAS,GAAc,SAAS,IAAI,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC;YAC/F,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;YAChD,MAAM,CAAC,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,UAAU,EAAE,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;YAC/E,OAAO;gBACL,IAAI,EAAE,IAAI;gBACV,IAAI,EAAE,CAAC,CAAC,MAAM,IAAI,EAAE;gBACpB,OAAO,EAAE,CAAC,CAAC,OAAO,IAAI,IAAI;gBAC1B,SAAS,EAAE,CAAC,CAAC,KAAK,IAAI,CAAC;gBACvB,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,EAAE;gBACnB,KAAK,EAAE,OAAQ,CAAwB,CAAC,KAAK,KAAK,QAAQ;oBACxD,CAAC,CAAE,CAAwB,CAAC,KAAM;oBAClC,CAAC,CAAC,IAAI;gBACR,WAAW,EAAE,CAAC,CAAC,KAAK,EAAE,WAAW,IAAI,CAAC;gBACtC,YAAY,EAAE,CAAC,CAAC,KAAK,EAAE,YAAY,IAAI,CAAC;gBACxC,gBAAgB,EAAE,CAAC,CAAC,KAAK,EAAE,gBAAgB,IAAI,CAAC;gBAChD,mBAAmB,EAAE,CAAC,CAAC,KAAK,EAAE,mBAAmB,IAAI,CAAC;aACvD,CAAC;QACJ,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC3D,8EAA8E;YAC9E,IAAI,CAAC,kCAAkC,CAAC,IAAI,CAAC,OAAO,CAAC;gBAAE,MAAM;QAC/D,CAAC;IACH,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,iBAAiB,EAAE,OAAO,EAAE,OAAO,IAAI,iBAAiB,EAAE,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC;AACjG,CAAC;AAED,SAAS,KAAK,CAAC,EAAU,IAAI,OAAO,IAAI,OAAO,CAAO,CAAC,CAAC,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC"}