CVE-2026-55673

PowSyBl (Power System Blocks) is a framework to build power system oriented software. Prior to 7.2.2, UnixLocalCommandExecutor and WindowsLocalCommandExecutor concatenate command arguments and environment variables into strings interpreted through bash -c or cmd /c without sufficient escaping. Attacker-controlled values reaching UnixLocalCommandExecutor.execute, WindowsLocalCommandExecutor.execute, LocalComputationManager.execute, ParallelLoadFlowActionSimulator.run, ActionSimulatorTool.run, AmplModelRunner.run, or AmplModelRunner.runAsync can break out of the intended command and execute arbitrary shell commands as the JVM user. The affected itools paths include action-simulator with task-count, security-analysis with external, and dynamic-security-analysis. Downstream CLI tools, libraries, REST front ends, and multi-tenant grid-analysis services that forward less-trusted contingency identifiers or computation parameters into these APIs can expose the injection remotely. This issue is fixed in version 7.2.2.

PowSyBl (Power System Blocks) is a framework to build power system oriented software. Prior to 7.2.2, UnixLocalCommandExecutor and WindowsLocalCommandExecutor concatenate command arguments and environment variables into strings interpreted through bash -c or cmd /c without sufficient escaping. Attacker-controlled values reaching UnixLocalCommandExecutor.execute, WindowsLocalCommandExecutor.execute, LocalComputationManager.execute, ParallelLoadFlowActionSimulator.run, ActionSimulatorTool.run, AmplModelRunner.run, or AmplModelRunner.runAsync can break out of the intended command and execute arbitrary shell commands as the JVM user. The affected itools paths include action-simulator with task-count, security-analysis with external, and dynamic-security-analysis. Downstream CLI tools, libraries, REST front ends, and multi-tenant grid-analysis services that forward less-trusted contingency identifiers or computation parameters into these APIs can expose the injection remotely. This issue is fixed in version 7.2.2.

Impact

Description

Both AbstractLocalCommandExecutor OS-dependent implementations are subject to CWE-78 (OS Command Injection), with a secondary CWE-88 (Argument Injection) concern via environment variables.

The local command executor build command strings via concatenation and executes them (through bash -c for Unix, or cmd /c for Windows). Any string argument or environment variable value reaching this executor can break out of the intended command and execute arbitrary shell code as the JVM user.

The sink is reachable through public APIs that accept List<String> arguments without any documented indication that elements undergo shell interpretation.

Any caller — CLI, library integration, or embedding service — that forwards data from a less-trusted source into these APIs inherits the vulnerability.

Vulnerable elements

The vulnerable elements are:

• Public methods:
  • UnixLocalCommandExecutor.execute(...)
  • WindowsLocalCommandExecutor.execute(...)
  • LocalComputationManager.execute(...)
  • ParallelLoadFlowActionSimulator.run(...)
  • ActionSimulatorTool.run(...)
  • AmplModelRunner.run(...)
  • AmplModelRunner.runAsync(...)
• itools commands:
  • action-simulator, when the task-count option is defined
  • security-analysis, when the external option is defined
  • dynamic-security-analysis

Characteristics

The vulnerability characteristics are the following:

• It allows arbitrary shell command execution as the JVM user (read, write, execute any file; spawn processes; exfiltrate data; etc.)
• There are multiple independent injection vectors within a single vulnerable call:
  • args elements (weakly escaped — bypassable via $(...), backticks, escaped quotes)
  • Environment variable values (unescaped — bypassable via ;, \n, or shell metacharacters)
• It is silent from the caller's perspective: the List<String> API gives no indication that shell interpretation occurs.
• It exposes downstream projects: any service embedding powsybl-core (REST front-ends, pypowsybl tooling, multi-tenant grid analysis platforms) that exposes contingency IDs or computation parameters to external input is exploitable without needing to touch powsybl's code directly.

Am I impacted?

You are vulnerable if you make direct calls to one of the vulnerable end-user methods or itools commands listed in the "Vulnerable elements" section, with user-provided parameters, without controlling them.

Patches

com.powsybl:powsybl-computation-local:7.2.2 and higher

Workarounds

If you cannot update your powsybl-computation-local version, you can check the content of the user-provided arguments when calling the vulnerable methods, by forbidding (if possible) the following characters:

• On Unix/Linux systems:
  • \, !, #, $, ^, &, *, (, ), {, }, |, [, ], \, ;, \, ", ,, <, >, ?,
• On Windows:
  • %, !, ", \n, \r, ^

PowSyBl (Power System Blocks) is a framework to build power system oriented software. Prior to 7.2.2, UnixLocalCommandExecutor and WindowsLocalCommandExecutor concatenate command arguments and environment variables into strings interpreted through bash -c or cmd /c without sufficient escaping. Attacker-controlled values reaching UnixLocalCommandExecutor.execute, WindowsLocalCommandExecutor.execute, LocalComputationManager.execute, ParallelLoadFlowActionSimulator.run, ActionSimulatorTool.run, AmplModelRunner.run, or AmplModelRunner.runAsync can break out of the intended command and execute arbitrary shell commands as the JVM user. The affected itools paths include action-simulator with task-count, security-analysis with external, and dynamic-security-analysis. Downstream CLI tools, libraries, REST front ends, and multi-tenant grid-analysis services that forward less-trusted contingency identifiers or computation parameters into these APIs can expose the injection remotely. This issue is fixed in version 7.2.2.

PowSyBl (Power System Blocks) is a framework to build power system oriented software. Prior to 7.2.2, UnixLocalCommandExecutor and WindowsLocalCommandExecutor concatenate command arguments and environment variables into strings interpreted through bash -c or cmd /c without sufficient escaping. Attacker-controlled values reaching UnixLocalCommandExecutor.execute, WindowsLocalCommandExecutor.execute, LocalComputationManager.execute, ParallelLoadFlowActionSimulator.run, ActionSimulatorTool.run, AmplModelRunner.run, or AmplModelRunner.runAsync can break out of the intended command and execute arbitrary shell commands as the JVM user. The affected itools paths include action-simulator with task-count, security-analysis with external, and dynamic-security-analysis. Downstream CLI tools, libraries, REST front ends, and multi-tenant grid-analysis services that forward less-trusted contingency identifiers or computation parameters into these APIs can expose the injection remotely. This issue is fixed in version 7.2.2.

Impact

Description

Both AbstractLocalCommandExecutor OS-dependent implementations are subject to CWE-78 (OS Command Injection), with a secondary CWE-88 (Argument Injection) concern via environment variables.

The local command executor build command strings via concatenation and executes them (through bash -c for Unix, or cmd /c for Windows). Any string argument or environment variable value reaching this executor can break out of the intended command and execute arbitrary shell code as the JVM user.

The sink is reachable through public APIs that accept List<String> arguments without any documented indication that elements undergo shell interpretation.

Any caller — CLI, library integration, or embedding service — that forwards data from a less-trusted source into these APIs inherits the vulnerability.

Vulnerable elements

The vulnerable elements are:

• Public methods:
  • UnixLocalCommandExecutor.execute(...)
  • WindowsLocalCommandExecutor.execute(...)
  • LocalComputationManager.execute(...)
  • ParallelLoadFlowActionSimulator.run(...)
  • ActionSimulatorTool.run(...)
  • AmplModelRunner.run(...)
  • AmplModelRunner.runAsync(...)
• itools commands:
  • action-simulator, when the task-count option is defined
  • security-analysis, when the external option is defined
  • dynamic-security-analysis

Characteristics

The vulnerability characteristics are the following:

• It allows arbitrary shell command execution as the JVM user (read, write, execute any file; spawn processes; exfiltrate data; etc.)
• There are multiple independent injection vectors within a single vulnerable call:
  • args elements (weakly escaped — bypassable via $(...), backticks, escaped quotes)
  • Environment variable values (unescaped — bypassable via ;, \n, or shell metacharacters)
• It is silent from the caller's perspective: the List<String> API gives no indication that shell interpretation occurs.
• It exposes downstream projects: any service embedding powsybl-core (REST front-ends, pypowsybl tooling, multi-tenant grid analysis platforms) that exposes contingency IDs or computation parameters to external input is exploitable without needing to touch powsybl's code directly.

Am I impacted?

You are vulnerable if you make direct calls to one of the vulnerable end-user methods or itools commands listed in the "Vulnerable elements" section, with user-provided parameters, without controlling them.

Patches

com.powsybl:powsybl-computation-local:7.2.2 and higher

Workarounds

If you cannot update your powsybl-computation-local version, you can check the content of the user-provided arguments when calling the vulnerable methods, by forbidding (if possible) the following characters:

• On Unix/Linux systems:
  • \, !, #, $, ^, &, *, (, ), {, }, |, [, ], \, ;, \, ", ,, <, >, ?,
• On Windows:
  • %, !, ", \n, \r, ^