verifiers 0.3.2.dev115__tar.gz → 0.3.2.dev116__tar.gz

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev115
+Version: 0.3.2.dev116
 Summary: Verifiers: Environments for LLM Reinforcement Learning
 Project-URL: Homepage, https://github.com/primeintellect-ai/verifiers
 Project-URL: Documentation, https://github.com/primeintellect-ai/verifiers

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/tests/v1/fixtures/echo_acp_resume_v1.py

@@ -53,13 +53,15 @@ class ACPResumeEnv(vf.SingleAgentEnv):
     async def run(self, task, agents):
         async with agents.agent.interaction(task) as interaction:
             first = await interaction.turn(
-                f"Remember the codeword {CODEWORD}. Reply with exactly READY."
+                f"Remember the codeword {CODEWORD} for this conversation only. "
+                "Do not call tools or store it in persistent memory. Reply with exactly READY."
             )
             segments = [first]
             if not first.terminated:
                 segments.append(
                     await interaction.turn(
-                        "Call the `recall` tool from the `resume` MCP server with the "
+                        "Use your tools or pre-imported Python skills to call `resume`'s "
+                        "`recall` with the "
                         "codeword from my previous message, then reply with exactly the "
                         "tool result."
                     )

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/browser_use/program.py

@@ -1,7 +1,7 @@
 # /// script
 # requires-python = ">=3.11"
 # dependencies = [
-#     "browser-harness==0.1.8",
+#     "browser-harness==0.1.13",
 #     "openai",
 #     "mcp==2.0.0",
 #     "httpx",

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/claude_code/harness.py

@@ -11,7 +11,7 @@ from verifiers.v1.trace import Trace
 
 CLAUDE_ACP_DIR = "/var/tmp/vf-claude-agent-acp-{version}-{acp_version}"
 PACKAGES_DIR = f"{CLAUDE_ACP_DIR}/packages"
-ACP_VERSION = "0.67.0"
+ACP_VERSION = "0.79.0"
 CLAUDE_BIN = f"{PACKAGES_DIR}/node_modules/.bin/claude"
 ACP_BIN = f"{PACKAGES_DIR}/node_modules/.bin/claude-agent-acp"
 CLAUDE_CONFIG_ROOT = ".vf-claude"
@@ -28,7 +28,7 @@ touch {ready}
 
 
 class ClaudeCodeHarnessConfig(HarnessConfig):
-    version: PinnedVersion = "2.1.232"
+    version: PinnedVersion = "2.1.278"
     """Claude Code release to install, pinned for reproducibility."""
 
 
@@ -79,6 +79,8 @@ class ClaudeCodeHarness(ACPHarness[ClaudeCodeHarnessConfig]):
                 "options": {
                     "strictMcpConfig": True,
                     "disallowedTools": self.config.disabled_tools or [],
+                    # A supplied title prevents the adapter's background title-model call.
+                    "extraArgs": {"name": trace.id},
                 }
             },
             **({"systemPrompt": {"append": system_prompt}} if system_prompt else {}),

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/codex/harness.py

@@ -19,7 +19,7 @@ logger = logging.getLogger(__name__)
 
 CODEX_DIR = "/var/tmp/vf-codex-{version}-{acp_version}"
 PACKAGES_DIR = f"{CODEX_DIR}/acp"
-ACP_VERSION = "1.2.0"
+ACP_VERSION = "1.12.0"
 CODEX_BIN = f"{PACKAGES_DIR}/node_modules/.bin/codex"
 ACP_BIN = f"{PACKAGES_DIR}/node_modules/.bin/codex-acp"
 INSTALL = r"""
@@ -35,7 +35,7 @@ touch {ready}
 
 
 class CodexHarnessConfig(HarnessConfig):
-    version: PinnedVersion = "0.147.0"
+    version: PinnedVersion = "0.155.1"
     """Codex release to install, pinned for reproducibility."""
     multi_agent: bool = False
     """Enable Codex's native multi-agent v2 tools."""

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/hermes_agent/harness.py

@@ -6,17 +6,28 @@ from pathlib import Path
 from verifiers.v1.acp import ACPConfig, ACPHarness
 from verifiers.v1.clients import ModelContext
 from verifiers.v1.configs.harness import HarnessConfig, PinnedVersion
-from verifiers.v1.harnesses.utils.install import remove_dir
+from verifiers.v1.harnesses.utils.install import ensure_installed, remove_dir
 from verifiers.v1.runtimes import Runtime
+from verifiers.v1.runtimes.base import _ENSURE_UV
 from verifiers.v1.task import TaskData
 from verifiers.v1.trace import Trace
 
 PROGRAM_SOURCE = (Path(__file__).resolve().parent / "program.py").read_text()
+HERMES_DIR = "/var/tmp/vf-hermes-agent-{version}"
+INSTALL = f"""
+set -e
+{_ENSURE_UV}
+command -v curl >/dev/null || (apt-get update -qq && apt-get install -y -qq curl ca-certificates >/dev/null)
+curl -fsSL "https://github.com/NousResearch/hermes-agent/archive/refs/tags/$VF_HERMES_VERSION.tar.gz" \\
+    | tar -xz --strip-components=1 -C "$VF_HERMES_DIR"
+uv sync --project "$VF_HERMES_DIR" --locked --no-dev --extra acp --extra mcp
+touch "$VF_HERMES_DIR/.ready"
+"""
 
 
 class HermesAgentHarnessConfig(HarnessConfig):
-    version: PinnedVersion = "0.19.0"
-    """Hermes Agent release to install, pinned for reproducibility."""
+    version: PinnedVersion = "v2026.9.14"
+    """Hermes Agent Git release tag to install, pinned for reproducibility."""
     use_bundled_skill: bool = False
     """Enable Hermes Agent's bundled skill catalog in addition to uploaded skills."""
 
@@ -27,9 +38,20 @@ class HermesAgentHarness(ACPHarness[HermesAgentHarnessConfig]):
     SUPPORTS_SKILLS = True
 
     async def setup(self, runtime: Runtime) -> None:
-        await runtime.prepare_uv_script(
-            PROGRAM_SOURCE.replace("{version}", self.config.version),
-            self.config.resolved_env,
+        # Hermes needs its source-tree assets and supports editable installs only.
+        directory = HERMES_DIR.format(version=self.config.version)
+        await ensure_installed(
+            runtime,
+            directory=directory,
+            ready=f"test -f {directory}/.ready",
+            install=INSTALL,
+            env={
+                **self.config.resolved_env,
+                "VF_HERMES_DIR": directory,
+                "VF_HERMES_VERSION": self.config.version,
+            },
+            label="Hermes Agent",
+            shell=("bash", "-o", "pipefail", "-c"),
         )
         await super().setup(runtime)
 
@@ -87,9 +109,12 @@ class HermesAgentHarness(ACPHarness[HermesAgentHarnessConfig]):
         system_prompt, prompt = self.resolve_prompt(data)
         return ACPConfig(
             env=env,
-            command=await runtime.prepare_uv_script(
-                PROGRAM_SOURCE.replace("{version}", self.config.version), env
-            ),
+            command=[
+                f"{HERMES_DIR.format(version=self.config.version)}/.venv/bin/python",
+                "-P",  # Keep task files from shadowing installed Hermes modules.
+                "-c",
+                PROGRAM_SOURCE,
+            ],
             prompt=prompt,
             system_prompt=system_prompt,
         )

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/hermes_agent/program.py

@@ -1,8 +1,8 @@
 # /// script
 # requires-python = ">=3.11,<3.14"
-# dependencies = ["hermes-agent[acp,mcp]=={version}"]
+# dependencies = []
 # ///
-"""Start Hermes Agent's native ACP server."""
+"""Start Hermes Agent's native ACP server in its installed project environment."""
 
 import os
 

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/kimi_code/harness.py

@@ -37,7 +37,7 @@ env \
 
 
 class KimiCodeHarnessConfig(HarnessConfig):
-    version: PinnedVersion = "0.36.0"
+    version: PinnedVersion = "2.0.2"
     """Kimi Code release to install, pinned for reproducibility."""
     transport: Literal["chat_completions", "responses", "anthropic_messages"] = (
         "chat_completions"

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/mini_swe_agent/program.py

@@ -1,6 +1,6 @@
 # /// script
 # requires-python = ">=3.10"
-# dependencies = ["mini-swe-agent=={version}", "litellm[proxy]==1.89.2"]
+# dependencies = ["mini-swe-agent=={version}", "litellm[proxy]==1.102.0"]
 # ///
 
 from minisweagent.run.mini import app

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/openclaw/harness.py

@@ -23,7 +23,7 @@ if [ ! -x "$VF_OPENCLAW_BIN" ]; then
     curl -fsSL --proto '=https' --tlsv1.2 https://openclaw.ai/install-cli.sh \
         | bash -s -- --prefix "$VF_OPENCLAW_DIR" --version "$VF_OPENCLAW_VERSION" --no-onboard
 fi
-set -- $(find "$VF_OPENCLAW_DIR/tools" -type f -path '*/openclaw/dist/session-accessor.sqlite-transcript-store-*.js')
+set -- $(find "$VF_OPENCLAW_DIR/tools" -type f -path '*/openclaw/dist/transcript-redact-*.mjs')
 [ "$#" -eq 1 ] || { echo "could not locate OpenClaw's transcript sanitizer for replay patch" >&2; exit 1; }
 "$VF_OPENCLAW_DIR/tools/node/bin/node" - "$1" <<'NODE'
 const fs = require("node:fs");
@@ -116,7 +116,7 @@ wait "$acp_pid"
 
 
 class OpenClawHarnessConfig(HarnessConfig):
-    version: PinnedVersion = "2026.8.1"
+    version: PinnedVersion = "2026.9.5"
     """OpenClaw release to install, pinned for reproducibility."""
     use_bundled_skill: bool = True
     """Enable OpenClaw's bundled skill catalog in addition to uploaded harness skills."""
@@ -180,6 +180,7 @@ class OpenClawHarness(ACPHarness[OpenClawHarnessConfig]):
                     "heartbeat": {"every": "0m"},
                     "sandbox": {"mode": "off"},
                     "model": {"primary": ctx.model},
+                    "utilityModel": "",  # Disable background recap model calls.
                 }
             },
             "tools": {

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/pi/harness.py

@@ -21,7 +21,7 @@ KEY_VAR = "PI_INTERCEPT_KEY"
 PI_DIR = "/var/tmp/vf-pi"
 PACKAGES_DIR = f"{PI_DIR}/mcp"
 PI_BIN = f"{PACKAGES_DIR}/node_modules/.bin/pi"
-MCP_VERSION = "2.28.0"
+MCP_VERSION = "2.34.0"
 ACP_VERSION = "0.0.33"
 MCP_ADAPTER = f"{PACKAGES_DIR}/node_modules/pi-mcp-adapter/index.ts"
 ACP_BIN = f"{PACKAGES_DIR}/node_modules/.bin/pi-acp"
@@ -44,7 +44,7 @@ fi
 
 
 class PiHarnessConfig(HarnessConfig):
-    version: PinnedVersion = "0.84.1"
+    version: PinnedVersion = "0.86.1"
     """Pi release to install, pinned for reproducibility."""
     transport: Literal["chat_completions", "responses", "anthropic_messages"] = (
         "chat_completions"

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/prime_agent/harness.py

@@ -19,10 +19,10 @@ logger = logging.getLogger(__name__)
 GITHUB_RELEASE_URL = (
     "https://github.com/PrimeIntellect-ai/prime-agent/releases/download"
 )
-PRIME_AGENT_COMMIT: Literal["81ae3cb34d27d38ee37f9e205a1e73694993b344"] = (
-    "81ae3cb34d27d38ee37f9e205a1e73694993b344"
+PRIME_AGENT_COMMIT: Literal["a7d791bc1be09793ed5f3ec05bf4cccbc60679ea"] = (
+    "a7d791bc1be09793ed5f3ec05bf4cccbc60679ea"
 )
-PRIME_AGENT_VERSION = "0.9.1"
+PRIME_AGENT_VERSION = "0.9.5"
 PRIME_AGENT_DIR = "/var/tmp/vf-prime-agent"
 STATE_ROOT = "/tmp/vf-prime-agent-runs"
 PROVIDER = "intercept"
@@ -50,10 +50,10 @@ for tarball in "$agent_tarball" "$ai_tarball" "$core_tarball" "$tui_tarball"; do
         "$release_url/$tarball" -o "$download_dir/$tarball"
 done
 printf '%s  %s\n' \
-    '573bce0cd004fc62052e9a924089941b7f39266ab71e66a94c85a1f9d35835ba' "$agent_tarball" \
-    '11b5b4cf67b6bb2d3420a44fb69181bc9d94d81e69a2b4fde07eb9c99f5faf4f' "$ai_tarball" \
-    'fb6f3a5dcc8b69c5eeb3beff722b5e0f09885c14849db50bc1d7c0f1d064151c' "$core_tarball" \
-    '4f3eaca2814944d3993073e0b88c0bde54a641ddb5132fbe107e392b997e38ec' "$tui_tarball" \
+    '349f1682c7909550842f1b04a71ba95814341b136474ade736df93f8ec006876' "$agent_tarball" \
+    '9ad0184b7b5f3d5c3b1677c0663335de32c092cbdb72697aeaf997f17910bb92' "$ai_tarball" \
+    '59aafeffb4b64eb997399d56a03f529b2f6604b228e5674372195277db52f04a' "$core_tarball" \
+    'e49f41170edfd1d0e72418729d2c117969ccf8df784dc55fe8c1a9d91778e103' "$tui_tarball" \
     > "$download_dir/SHA256SUMS"
 (cd "$download_dir" && sha256sum -c SHA256SUMS)
 mkdir "$download_dir/package-root"
@@ -87,8 +87,8 @@ PRIME_AGENT_BOOTSTRAP_TOOLS_ON_INSTALL=1 npm install -g \
 
 
 class PrimeAgentHarnessConfig(HarnessConfig):
-    commit: Literal["81ae3cb34d27d38ee37f9e205a1e73694993b344"] = PRIME_AGENT_COMMIT
-    """Prime Agent main commit to install."""
+    commit: Literal["a7d791bc1be09793ed5f3ec05bf4cccbc60679ea"] = PRIME_AGENT_COMMIT
+    """Prime Agent release commit to install."""
 
     autonomous: bool = False
     """Enable Prime Agent's autonomous continuation loop."""

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/rlm/harness.py

@@ -69,7 +69,7 @@ class CompactionConfig(BaseConfig):
 
 class RLMHarnessConfig(HarnessConfig):
     version: str = Field(
-        default="00d8aa713c820023033bce6aa2eacea6c37e675d", min_length=1
+        default="fc5734d20c5fe74f6ab527ecb7ab10299f3bb965", min_length=1
     )
     """Git ref (branch, tag, or commit) of nano-rlm to install. Must know every
     field this harness puts on the wire, i.e. be at least the default ref."""

{verifiers-0.3.2.dev115 → verifiers-0.3.2.dev116}/verifiers/v1/harnesses/terminus_2/harness.py

@@ -13,7 +13,7 @@ logger = logging.getLogger(__name__)
 
 
 class Terminus2HarnessConfig(HarnessConfig):
-    version: PinnedVersion = "0.21.0"
+    version: PinnedVersion = "0.23.0"
     """Harbor release to install, pinned for reproducibility."""