recourse-cli

Parses Terraform execution plans (JSON format) to identify resource changes, dependencies, and potential blast radius. Analyzes which resources will be created, modified, or destroyed and traces downstream impacts through resource dependency graphs. Integrates with MCP protocol to expose analysis results to AI agents for decision-making before apply operations.

Analyzes shell commands (bash, sh, zsh) to predict side effects including file system mutations, process spawning, network calls, and environment variable changes. Uses pattern matching and AST-like parsing to identify dangerous operations (rm, dd, curl with sudo, etc.) and traces command chains to estimate overall system impact. Exposes findings through MCP tool interface for agent evaluation.

Validates MCP tool calls against their schemas and predicts consequences of tool execution based on tool metadata and parameter values. Analyzes tool definitions to identify which tools perform mutations, access sensitive resources, or have side effects. Evaluates whether a proposed tool call aligns with agent intent and flags potentially dangerous parameter combinations (e.g., delete with wildcard patterns).

Builds and traverses dependency graphs from Terraform plans and MCP tool definitions to trace resource relationships and impact chains. Identifies direct dependencies (explicit resource references) and estimates transitive impacts when resources are modified or deleted. Generates visual or textual representations of dependency chains to help agents understand cascading effects.

Assigns risk scores and severity classifications to proposed actions (Terraform changes, shell commands, tool calls) based on impact type, blast radius, and resource criticality. Uses a scoring model that considers factors like number of affected resources, whether changes are reversible, and whether critical infrastructure is involved. Provides severity labels (low, medium, high, critical) to help agents make informed decisions.

Implements a Model Context Protocol (MCP) server that exposes consequence analysis capabilities as MCP tools callable by AI agents. Handles MCP protocol communication, tool registration, parameter marshaling, and result serialization. Allows agents to invoke consequence analysis tools through standard MCP client interfaces without direct library imports.

Parses multiple input formats including Terraform JSON plans, shell command text, and MCP tool definition schemas. Uses format-specific parsers to extract relevant information (resource changes, command operations, tool metadata) and normalize into internal representations for analysis. Handles format variations and provides clear error messages for malformed inputs.

Analyzes operations to determine whether changes are reversible and identifies operations that could cause permanent data loss. Classifies operations as reversible (can be undone via backup/rollback), partially reversible (some data recoverable), or irreversible (permanent loss). Detects high-risk patterns like database deletions, encryption key destruction, and unbackup'd resource removal.