@bivy/bivy 0.16.25-staging.23 → 0.16.25-staging.24

package/bin/agent-manifest.json

@@ -22,7 +22,7 @@
       "hidden": false,
       "supportTier": "supported",
       "certification": "release-tested",
-      "testedVersion": "1.18.30",
+      "testedVersion": "1.18.31",
       "headlessFlags": [
         "run",
         "-s"

package/dist/agents/claude-code/integration.js

@@ -4,13 +4,21 @@ import { defineAgentIntegration } from "../definition.js";
 import { withExactCapabilitySurface } from "../../runtime/types.js";
 import { createCredentialStore } from "../../runtime/credentials.js";
 import { ClaudeCodeRuntime, claudeRuntimeFromEnv, claudeSdkInstalled, } from "./runtime.js";
-export const CLAUDE_TESTED_VERSION = "0.3.270";
+export const CLAUDE_TESTED_VERSION = "0.3.278";
 const CLAUDE_CAPABILITIES = withExactCapabilitySurface({
     toolInterception: true,
     modelSelection: true,
     packages: false,
     resume: true,
     fork: true,
+    // Must mirror ClaudeCodeRuntime.capabilities (runtime.ts): the catalog is a
+    // consumed source of truth (the fork-fidelity matrix and any fork-target
+    // fidelity preview read these), so under-reporting here makes Claude look
+    // like a seeded-only fork target when its runtime actually delivers a
+    // byte-exact same-runtime fork (forkTransport) and a full cross-runtime
+    // transcript replay INTO it (forkHistoryImport, via importHistoryForFork).
+    forkTransport: true,
+    forkHistoryImport: true,
     usageReporting: true,
     nativeSessionDiscovery: true,
     nativeSessionAdoption: true,

package/dist/agents/codex/integration.js

@@ -9,7 +9,7 @@ import { codexCredentialPreflight } from "../../runtime/codex-preflight.js";
 import { deleteCodexSession, discoverNativeCodexSessions, loadCodexTranscript, writeCodexRollout, exportCodexRollout, importCodexRollout, } from "../../runtime/codex-sessions.js";
 import { ProtocolRuntime } from "../../runtime/protocol.js";
 import { codexSlashCommands } from "../../runtime/slash-commands.js";
-export const CODEX_TESTED_VERSION = "0.154.0";
+export const CODEX_TESTED_VERSION = "0.155.1";
 const CODEX_AVAILABLE_CACHE = new Map();
 function codexCommand() {
     return process.env.BIVY_CODEX_BIN?.trim() || "codex";
@@ -146,6 +146,14 @@ export function codexIntegration(origin) {
                     resume: true,
                     packages: false,
                     fork: true,
+                    // Mirror the ProtocolRuntime this integration builds (createCodexRuntime
+                    // above wires exportForFork+importForFork and writeHistory, which flip
+                    // capabilities.forkTransport/forkHistoryImport on). The catalog is a
+                    // consumed source of truth (fork-fidelity matrix / fork-target preview),
+                    // so omitting these made Codex look seeded-only when its runtime does a
+                    // byte-exact codex→codex fork and a full transcript replay INTO it.
+                    forkTransport: true,
+                    forkHistoryImport: true,
                     usageReporting: true,
                     interactiveTui: installed,
                     nativeSessionDiscovery: true,

package/dist/agents/pi/integration.js

@@ -5,7 +5,7 @@ import os from "node:os";
 import path from "node:path";
 import { defineAgentIntegration } from "../definition.js";
 import { PI_CAPABILITIES } from "./capabilities.js";
-export const PI_TESTED_VERSION = "0.85.1";
+export const PI_TESTED_VERSION = "0.86.1";
 export function piAgentDir() {
     return process.env.PI_CODING_AGENT_DIR?.trim() || path.join(os.homedir(), ".pi", "agent");
 }

package/dist/agents/profiles.js

@@ -43,7 +43,7 @@ export const AGENT_PROFILES = {
         // Approve/Deny + session/load resume + a real model picker), the same bar Pi,
         // Claude Code, and Codex clear. See `acp` below for the version fallback.
         supportTier: "supported",
-        testedVersion: "1.18.30",
+        testedVersion: "1.18.31",
         blurb: "The most widely used open-source coding harness (OpenCode CLI).",
         // `opencode run -s <id> "<prompt>"` continues a prior session by its own id
         // (`-s, --session  session id to continue`, per `opencode run --help`).

package/dist/certification/generated.js

@@ -3,9 +3,9 @@
 export const CERTIFICATION_MATRIX = {
     schemaVersion: 1,
     agents: [
-        { id: "claude-code-sdk", status: "active", executionMode: "protocol", pinnedVersion: "0.3.270", capabilities: ["toolInterception", "modelSelection", "resume"] },
-        { id: "codex-approvals", status: "active", executionMode: "protocol", pinnedVersion: "0.154.0", capabilities: ["toolInterception", "modelSelection", "resume"] },
-        { id: "pi", status: "active", executionMode: "protocol", pinnedVersion: "0.85.1", capabilities: ["toolInterception", "modelSelection", "resume"] },
-        { id: "opencode", status: "active", executionMode: "protocol", pinnedVersion: "1.18.30", capabilities: ["toolInterception", "modelSelection", "resume"] },
+        { id: "claude-code-sdk", status: "active", executionMode: "protocol", pinnedVersion: "0.3.278", capabilities: ["toolInterception", "modelSelection", "resume"] },
+        { id: "codex-approvals", status: "active", executionMode: "protocol", pinnedVersion: "0.155.1", capabilities: ["toolInterception", "modelSelection", "resume"] },
+        { id: "pi", status: "active", executionMode: "protocol", pinnedVersion: "0.86.1", capabilities: ["toolInterception", "modelSelection", "resume"] },
+        { id: "opencode", status: "active", executionMode: "protocol", pinnedVersion: "1.18.31", capabilities: ["toolInterception", "modelSelection", "resume"] },
     ]
 };

package/dist/runtime/cli-parsers.js

@@ -696,6 +696,25 @@ function rawExitCode(raw) {
     const exit = o.exit_code ?? o.exitCode ?? o.code;
     return typeof exit === "number" ? exit : undefined;
 }
+/** Decode a UTF-8 byte array carried as a JSON `number[]` (Grok's
+ * `rawOutput.output` transports command output this way). Returns the decoded
+ * string, or undefined when the value isn't a plausible byte array. Used only as
+ * a LAST resort — a clean ACP `content` block is always preferred — so a result
+ * frame with no display content still renders readable text instead of a raw
+ * number array (`[97,46,116,120,116]`). */
+function decodeByteArray(v) {
+    if (!Array.isArray(v) || v.length === 0)
+        return undefined;
+    if (!v.every((n) => typeof n === "number" && Number.isInteger(n) && n >= 0 && n <= 255))
+        return undefined;
+    try {
+        const text = Buffer.from(v).toString("utf8");
+        return text.includes("\uFFFD") ? undefined : text;
+    }
+    catch {
+        return undefined;
+    }
+}
 /** Extract an ACP-style tool update. Two transports converge here:
  *   1. A JSON-RPC `session/update` notification (`params.update`), the true ACP
  *      wire form used by Gemini/Qwen/Copilot etc.
@@ -735,8 +754,16 @@ function acpToolUpdate(msg) {
         // the exit code from the raw payload when we take the content text instead.
         const contentText = flattenAcpContent(update.content);
         const exit = rawExitCode(raw);
-        const output = contentText !== undefined
-            ? (exit !== undefined ? { content: contentText, exit_code: exit } : contentText)
+        // Last-resort text when no clean ACP `content` block is present: decode a
+        // byte-array payload (Grok's rawOutput.output) so the card shows readable
+        // output instead of a raw `[97,46,…]` number array. Never overrides a real
+        // content block.
+        const decoded = contentText === undefined
+            ? (decodeByteArray(raw) ?? (raw && typeof raw === "object" && !Array.isArray(raw) ? decodeByteArray(raw.output) : undefined))
+            : undefined;
+        const text = contentText ?? decoded;
+        const output = text !== undefined
+            ? (exit !== undefined ? { content: text, exit_code: exit } : text)
             : (raw ?? update.content);
         return { kind: "result", id, output, error: status === "failed" || status === "error" };
     }

package/dist/runtime/index.js

@@ -53,6 +53,7 @@ import { ProtocolRuntime, protocolRuntimeFromEnv, protocolCommandsFromEnv } from
 import { codexSlashCommands, opencodeSlashCommands } from "./slash-commands.js";
 import { withExactCapabilitySurface } from "./types.js";
 import { installedAgentContributions } from "../plugins/store.js";
+import { forkCapsFromInfo } from "../session/fork-matrix.js";
 import { currentBivyVersion } from "../app-version.js";
 const PREFLIGHT_BEHAVIORS = {
     codex: (env) => codexCredentialPreflight(env),
@@ -552,6 +553,17 @@ function cliAgentInfo(id, spec) {
     }
     const authOwner = spec.authOwner ?? "agent";
     const credentialProviders = [...new Set((spec.model?.models ?? []).flatMap((model) => model.provider ? [model.provider] : []))];
+    // Fork fidelity, mirrored from what makeCliRuntime actually wires up. An
+    // ACP-promoted agent with a portable session store (OpenCode's SQLite) gets
+    // export/import + writeHistory hooks in acpRuntimeOptions, so its runtime
+    // advertises a byte-exact same-runtime fork (forkTransport) AND a full
+    // cross-runtime transcript replay INTO it (forkHistoryImport). Reporting a
+    // blanket fork:false here hid a full-fidelity fork target from the catalog
+    // and the generated fork matrix. Gated on the shipped default (acp.preferred)
+    // rather than a live binary probe so the capability is install-independent —
+    // the honest "what this adapter does by default", matching how resume/model
+    // are described before the optional BIVY_AGENT_PROBE refinement.
+    const nativeStoreFork = spec.behaviors?.sessionStore === "opencode" && spec.acp?.preferred === true;
     return {
         id,
         executionMode,
@@ -570,7 +582,8 @@ function cliAgentInfo(id, spec) {
             modelSelection,
             resume,
             packages: false,
-            fork: false,
+            fork: nativeStoreFork,
+            ...(nativeStoreFork ? { forkTransport: true, forkHistoryImport: true } : {}),
             usageReporting,
             // Any agent with a readable on-disk session store (Codex, OpenCode) or
             // native-session behavior (Grok) can locate the session a `bivy run`
@@ -1009,6 +1022,22 @@ function agentRegistry(pluginDataDir) {
 export function listRegisteredAgents() {
     return agentRegistry().list().map((entry) => applyCertification(entry.describe()));
 }
+// Meta/host rows that are not concrete coding agents (they describe the generic
+// transports themselves), plus the hidden plain-`codex` exec runtime, which is a
+// duplicate of the governed `codex-approvals` we surface as "Codex".
+const FORK_MATRIX_EXCLUDE = new Set(["generic-cli", "bivy-agent-protocol", "acp", "codex"]);
+/** The built-in coding agents and their fork capabilities, derived from the live
+ *  registry, for docs/fork-matrix.md and its drift-guard test. Restricted to the
+ *  maintained (distribution) integrations so the generated doc is deterministic
+ *  regardless of which agents happen to be installed or which plugins are present. */
+export function forkMatrixAgents() {
+    return listRegisteredAgents()
+        .filter((info) => info.status !== "planned" &&
+        !FORK_MATRIX_EXCLUDE.has(info.id) &&
+        info.source?.kind === "package" &&
+        info.source.location === "distribution")
+        .map(forkCapsFromInfo);
+}
 /** Resolve an id or historical alias through the authoritative registry. */
 export function canonicalAgentId(id) {
     return agentRegistry().get(id.toLowerCase())?.id;

package/dist/session/fork-matrix.js

@@ -28,6 +28,18 @@ export function forkTier(source, dest) {
         return "replayed";
     return "seeded";
 }
+/** Project a registry RuntimeInfo onto the two flags that decide fork fidelity.
+ *  Keeps the registry the single source of truth: an agent that declares its
+ *  real fork capabilities (see the maintained integrations + cliAgentInfo)
+ *  appears with the correct tier automatically. */
+export function forkCapsFromInfo(info) {
+    return {
+        id: info.id,
+        ...(info.displayName ? { displayName: info.displayName } : {}),
+        forkTransport: info.capabilities.forkTransport === true,
+        forkHistoryImport: info.capabilities.forkHistoryImport === true,
+    };
+}
 /** Every ordered (source, dest) pair and its tier. */
 export function forkMatrix(agents) {
     const cells = [];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bivy/bivy",
-  "version": "0.16.25-staging.23",
+  "version": "0.16.25-staging.24",
   "type": "module",
   "license": "AGPL-3.0-only",
   "description": "Run coding agents on machines you own. Open-source, self-hostable agent workspace.",
@@ -51,9 +51,9 @@
     "zod": "^4.0.0"
   },
   "optionalDependencies": {
-    "@anthropic-ai/claude-agent-sdk": "0.3.270",
-    "@earendil-works/pi-ai": "0.85.1",
-    "@earendil-works/pi-coding-agent": "0.85.1"
+    "@anthropic-ai/claude-agent-sdk": "0.3.278",
+    "@earendil-works/pi-ai": "0.86.1",
+    "@earendil-works/pi-coding-agent": "0.86.1"
   },
   "overrides": {
     "fast-uri": "3.1.6",