The AI Toolchain — issue -048, July 2, 2026

LocalFileSystemTools gains file-read support and directory confinement controls via new flags.

$ git clone --branch v2.6.21 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:
$ git checkout v2.6.21

from agno.tools.local_file_system import LocalFileSystemTools

tools = LocalFileSystemTools(
    target_directory="/data/agent-workspace",
    enable_read_file=True,
    restrict_to_base_dir=True,
)

• ›Adds enable_read_file flag to LocalFileSystemTools to expose a read-file tool to agents.
• ›Adds restrict_to_base_dir flag to LocalFileSystemTools to confine all file operations within target_directory by default; set restrict_to_base_dir=False to opt out.

PydanticAI v2.3.0 adds a native Z.AI (Zhipu AI) provider with thinking support.

$ git clone --branch v2.3.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v2.3.0

• ›Adds native Z.AI (Zhipu AI) provider with thinking support.

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

Claude Code v2.1.199 adds stacked skill invocations, auto-retry on 429s, and an elevated retry watchdog cap.

$ git clone --branch v2.1.199 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:
$ git checkout v2.1.199

$ /skill-security /skill-codestyle review this PR for vulnerabilities and style issues

• ›Supports stacked slash-skill invocations (/skill-a /skill-b do XYZ) loading up to 5 leading skills simultaneously, not just the first.
• ›Transient server rate-limit errors (429s unrelated to usage limits) are now automatically retried with backoff for subscribers instead of failing the turn.
• ›New CLAUDE_CODE_RETRY_WATCHDOG behaviour raises the default retry count for non-capacity transient errors to 300 and removes the previous cap of 15 on CLAUDE_CODE_MAX_RETRIES.
• ›Pull-request links in claude agents session rows now render as bare #N for cleaner output.

an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM

Goose v1.40.0 adds goose review, new slash commands, a hooks system, 10+ new AI providers, and OTLP-based cross-tool detection.

$ git clone --branch v1.40.0 https://github.com/block/goose.git
# already have the repo? check out this version:
$ git checkout v1.40.0

$ goose review

$ /model

$ GOOSE_MAX_TOOL_RESPONSE_SIZE=51200 goose run --text "Summarize the logs in ./logs/"

• ›Adds goose review command for local AI-powered code review.
• ›New /status slash command in the CLI to check session state.
• ›New /model slash command for switching models mid-session.
• ›New /goal slash command for agent self-evaluation before finishing a task.
• ›Introduces a hooks system with PreToolUse denial support and blocking Stop hook decisions for extensibility and policy enforcement.

Qwen Code v0.19.5 adds HTTPS/TLS serving, a /effort command, DingTalk proactive send, and credential redaction for worker output.

$ git clone --branch v0.19.5 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.19.5

$ qwen serve --tls-cert /etc/certs/server.crt --tls-key /etc/certs/server.key

$ /effort low

• ›Adds --tls-cert and --tls-key flags to the serve subcommand to enable HTTPS/TLS on the server.
• ›Adds the /effort slash command for unified control of reasoning effort level from the CLI.
• ›Adds listSessions to ChannelAgentBridge for programmatic session enumeration in channel integrations.
• ›Adds credential redaction for worker stderr forwarding, preventing secrets from leaking into logs.
• ›Adds DingTalk proactive send support for channel loops, enabling outbound messaging from agent sessions.

llama.cpp b9860 adds llama_ftype_name() to the public C API for querying model quantization type names.

$ git clone --branch b9860 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b9860

• ›Adds llama_ftype_name() to the public C API, returning a human-readable quantization name (e.g. Q8_0, Q4_K - Medium) for a given llama_ftype enum value; the returned pointer is valid for the lifetime of the model and allocation-free.

LanceDB v0.34.0 adds FM-Index substring search, table branches, OAuth support, Polars integration, and approx vector query mode.

$ git clone --branch python-v0.34.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.34.0

results = (
    table.search(query_vector)
    .where(Expr.field("category").isin(["malware", "phishing", "ransomware"]))
    .limit(10)
    .to_pandas()
)

results = table.search(query_vector).approx(True).limit(50).to_pandas()

• ›Adds update_field_metadata method to edit per-field Arrow metadata on table columns.
• ›Adds isin support to the Expr builder for filter expressions.
• ›Adds approx mode to vector queries, letting callers explicitly request approximate nearest-neighbor search.
• ›Adds FM-Index scalar index type for substring search via create_index.
• ›Adds table branch support to local and remote tables and Python/TypeScript bindings, including checking out a specific version on a branch.

$ git clone --branch v0.31.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.31.0

from lancedb.query import Expr
tbl.delete(Expr.field("status").isin(["stale", "archived"]))

$ git clone --branch python-v0.34.0-beta.6 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.34.0-beta.6

$ git clone --branch v0.31.0-beta.6 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.31.0-beta.6

Langfuse v3.203.2 adds agent skills, auto-rename conversations, and an event-propagation health check.

$ git clone --branch v3.203.2 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.203.2

• ›Adds Langfuse skills support to the agent, enabling the agent to invoke built-in capabilities.
• ›Adds auto-rename for agent conversations based on context.
• ›Adds event-propagation stuck health check to the worker for monitoring pipeline reliability.
• ›Adds Salesforce org ID, user auth methods, and SSO domain data to core data export.

Weave v0.53.1 expands the TypeScript agent API with Turn.record(), SubAgent.record(), richer LLM.record() fields, and adds predict-only cost totals to eval summaries.

$ git clone --branch v0.53.1 https://github.com/wandb/weave.git
# already have the repo? check out this version:
$ git checkout v0.53.1

llm.record({
  mediaAttachments: [...],
  responseId: 'resp_abc123',
  responseModel: 'gpt-4o',
  finishReasons: ['stop'],
  outputType: 'text'
});

const subagent = new Subagent({
  systemInstructions: 'You handle tool calls only.'
});
await subagent.record({ ... });

• ›Adds userMessage and systemInstructions parameters when creating Turns in the TypeScript API.
• ›Adds agent_id, agent_description, and agent_version parameters when creating Turns in the TypeScript API.
• ›Adds agent_id, agent_description, and agent_version parameters when creating Conversations in the TypeScript API.
• ›Adds systemInstructions parameter when creating Subagents in the TypeScript API.
• ›Adds systemInstructions parameter when creating LLMs in the TypeScript API.

$ git clone --branch v0.53.0 https://github.com/wandb/weave.git
# already have the repo? check out this version:
$ git checkout v0.53.0

import { init } from 'weave';
await init({ projectName: 'my-project', useOTelv2: true });