Stable API Reference¶

ABI exposes one transport-neutral lifecycle through the CLI, MCP server, HTTP jobs, and Python. Agent integrations should prefer JSON envelopes and must not parse human-readable terminal output.

Agent Boundary¶

Every agent-facing operation returns one of three statuses:

success: The result field contains the structured payload.
confirmation_required: Execution is waiting for explicit user approval.
error: Inspect error_code and diagnostic_hints before retrying.

The safe lifecycle is list-types → plan → dry-run → inspect → run → report. All CLI lifecycle commands accept --output-json. See Agent Usage Guide for MCP configuration, exported tool descriptors, permissions, examples, and recovery rules.

Python Entry Points¶

The supported Python package entry points are:

abi.get_agent_guide(): Return a compact operating guide suitable for an Agent system prompt.
abi.list_plugins_summary(): Return installed plugin metadata without starting a workflow.
abi.agent.ABIAgentInterface: Transport-neutral implementation behind CLI JSON, MCP, and HTTP adapters.
abi.plugins.list_plugins() and abi.plugins.get_plugin(): Discover or load registered analysis plugins.

Example:

import json

from abi.agent import ABIAgentInterface

interface = ABIAgentInterface()
envelope = json.loads(interface.list_types())
if envelope["status"] == "success":
    print(envelope["result"])

Execution methods require the same explicit confirmation as the CLI. Do not bypass the lifecycle or call plugin internals from an Agent integration.

Developer Interfaces¶

Plugin authors implement the protocols in abi.interfaces and declare tools, DAGs, schemas, and report metadata in their plugin directory. The maintained contracts and examples live in:

Plugin Development Guide — plugin protocols, manifests, and tests.
Workflow Validation and Scientific Evidence Plan — biological and software validation boundaries.
Release-ready runtime locks — release-ready tools and resource certification.
ABI Scientific Figure Compiler (abi-sciplot) — validated scientific figure specifications.
Multi-LLM Tool Descriptor Export — provider descriptor export formats.

The source repository remains authoritative for concrete Python signatures. Use the version shown in the documentation announcement when selecting source for a deployed documentation build.