Tools Registry¶

CONCEPT:AU-ECO.messaging.native-backend-abstraction — Agent Tool System

Overview¶

The tools/ package provides the PydanticAI tool surface used by direct agent execution. register_agent_tools is the authoritative registry; optional groups are enabled through AgentConfig, and graph orchestrators expose only execute_graph, the protocol-agnostic graph execution authority, so routing remains isolated.

Tool Categories¶

Agent & Team Tools¶

Module	Key Functions	Description
`agent_tools`	`share_reasoning`, `invoke_specialized_agent`	Core agent capabilities
`team_tools`	`spawn_team`, `assign_team_task`, `message_teammate`	Multi-agent coordination
`a2a_tools`	`list_a2a_peers`, `register_a2a_peer`, `delete_a2a_peer`	Agent-to-Agent peer registry

Developer Tools¶

Module	Key Functions	Description
`developer_tools`	`project_search` plus shared KG tools	Read-only code and knowledge discovery
`git_tools`	`get_git_status`, `list_worktrees`	Privacy-safe repository inspection
`workspace_tools`	`read_workspace_file`, `get_skill_content`, `list_files`	Bounded, read-only workspace navigation

Knowledge & Memory Tools¶

Module	Key Functions	Description
`kg_tools`	`kg_search`, `kg_recall`, `kg_query`	Shared graph discovery and queries
`knowledge_tools`	memory CRUD, SDD synchronization, and governed knowledge-base operations	Knowledge and memory lifecycle
`memory_tools`	`read_agents_md`	Bounded, sanitized project instructions
`kg_evolution_tools`	`extract_and_ingest_triples`	Governed triple extraction and ingestion
`kg_share_tools`	`export_subgraph`, `import_agent_card`	KG data sharing
`pattern_tools`	`run_manual_test`	Pattern test execution

Module	Key Functions	Description
`x_search_tool`	`x_search`, `browse_x_post`	Native X search and post browsing using xAI authentication

Infrastructure Tools¶

Module	Key Functions	Description
`scheduler_tools`	`schedule_task`, `list_tasks`, `delete_task`, `view_cron_log`	Task scheduling
`mcp_sync_tool`	`trigger_mcp_sync`	MCP server discovery
`sdd_tools`	project context, specification, task, import/export, and TDD operations	Spec-driven development
`self_improvement_tools`	improvement cycles, skill proposals, and experiment queries	Self-improvement
`style_tools`	`set_output_style`, `list_output_styles`	Output-style selection
`onboarding_tools`	`bootstrap_project`	Project onboarding

Tool Registration¶

Tools are registered with the agent during creation:

from agent_utilities.agent.factory import create_agent

# Tools are automatically loaded based on skill_types
agent = create_agent(
    name="MyAgent",
    skill_types=["universal", "graphs"],
)

Tool Guard (`security/tool_guard.py`)¶

Safety layer that validates tool calls before execution: - Allowlist/blocklist: Restrict which tools are available - Rate limiting: Prevent excessive tool calls - Approval gates: Require human approval for destructive operations

Tool Filtering (`tool_filtering.py`)¶

Dynamic tool selection based on context: - Reduces token usage by only exposing relevant tools - Uses semantic similarity to match tools to the current task - Supports explicit tool pinning via configuration

Tool Registry (`tools/tool_registry.py`)¶

Central entry point that wires all available tool modules onto a PydanticAI agent at creation time:

from agent_utilities.tools.tool_registry import register_agent_tools

# Registers the relevant tool modules onto the agent, guarded against duplicate
# registration and the configured optional-tool policy.
register_agent_tools(agent, graph_bundle=my_graph_bundle)

X & xAI Integration¶

The x_search_tool provides native capabilities to search X posts and browse individual posts directly from an agent. It use the secure XaiAuthManager to obtain OAuth 2.0 access tokens.

Functions¶

x_search(query: str, allowed_x_handles=None, excluded_x_handles=None, from_date="", to_date=""): Searches X posts for a query and returns matching posts, authors, and text. Exactly one of allowed_x_handles/excluded_x_handles may be set.
browse_x_post(url: str, auto_ingest: bool = False): Fetches the text, author, and engagement metrics for a specific X post by its URL; optionally classifies and ingests it into the KG.

Example Code¶

# The agent will automatically call these tools when given an X URL or asked to search X:
result = browse_x_post("https://x.com/gkisokay/status/2056726149074657704")
print(result)