Improper Control of Generation of Code ('Code Injection')

PraisonAI before 1.6.78 contains a remote code execution vulnerability in CodeAgent._execute_python() that executes LLM-generated Python code without AST validation, import restrictions, or sandbox enforcement. Attackers can influence LLM output through prompt injection to exfiltrate all environment secrets and execute arbitrary code on the host system.Show less

The WP Ultimate CSV Importer – WordPress Import & Export for CSV, XML & Excel plugin for WordPress is vulnerable to Remote Code Execution in all versions up to, and including, 8.0.1 via the 'MappedFields' parameter. This is due to missing capability checks on the AJAX handlers for install_addon, saveMappedFields, and StartImport, combined with the plugin nonce being exposed to any authenticated user who can load an admin page, allowing a Subscriber to install the Import WooCommerce add-on, persist attacker-controlled PHP expressions in the MappedFields parameter, and trigger evaluation via eval() in ImportHelpers::get_meta_values(). This makes it possible for authenticated attackers, with subscriber-level access and above, to execute code on the server.Show less

Grav before 2.0.2 contains a Twig sandbox bypass that allows a page author (any admin.pages user, or anyone able to write to user/pages) to exfiltrate configuration secrets. Although the sandbox replaces the 'config' variable with a redacted facade and strips Config::get/toArray from the method allowlist, the raw container remains accessible via the allow-listed grav.offsetGet('config'), which returns the real Config object. Allow-listed object-dumping filters (json_encode, print_r, yaml_encode) then serialize that object at the PHP level without invoking the sandbox method gate, exposing the full config tree including plugin secrets such as SMTP credentials, API keys, and plugin DB credentials. This is an incomplete fix for GHSA-j274-39qw-32c9.Show less

PraisonAI versions before 4.6.78 contain a code injection vulnerability in deploy/api.py where the agents_file parameter is directly interpolated into an f-string without sanitization. Attackers can inject arbitrary Python code that executes when the generated server code runs via subprocess.Popen().Show less

A vulnerability was found in MyEMS up to 6.4.0. The affected element is the function on_post of the file myems-api/core/svg.py of the component Admin Backend. The manipulation of the argument new_values['data'] results in cross site scripting. The attack can be launched remotely. The exploit has been made public and could be used. Upgrading to version 6.5.0 is sufficient to fix this issue. The patch is identified as 4a97edfbd786c779d0322054833b21ddf54d5b06. It is suggested to upgrade the affected component. The issue report remains open even though there is an official fix for it.Show less

Langroid is a framework for building large-language-model-powered applications. Versions prior to 0.65.2 are vulnerable to a critical Sandbox Escape leading to Remote Code Execution (RCE) in its `TableChatAgent` and `VectorStore` capabilities. When these agents evaluate LLM-generated tool messages with `full_eval=True`, they attempt to sandbox the execution by explicitly setting `locals` to an empty dictionary `{}` inside Python's `eval()` function. However, this relies on an incomplete understanding of Python's execution model. Because `__builtins__` is not explicitly scrubbed from the `globals` dictionary mapping, Python implicitly injects all built-ins during execution, granting full access to functions like `__import__('os').system()`. Since `TableChatAgent.pandas_eval()` executes external LLM outputs natively, this bypass permits any attacker providing prompt payload to achieve unauthenticated RCE on the host system. Version 0.65.2 patches the issue.Show less

A vulnerability was identified in NousResearch hermes-agent up to 2026.5.29.2. Affected by this issue is the function MatrixAdapter._markdown_to_html of the file gateway/platforms/matrix.py of the component Matrix Adapter. Such manipulation leads to cross site scripting. The attack can be executed remotely. The exploit is publicly available and might be used. The pull request to fix this issue awaits acceptance.Show less

Vim is an open source, command line text editor. Prior to 9.2.0735, the C omni-completion script in runtime/autoload/ccomplete.vim interpolates the typeref: or typename: extension field of a tags entry, without escaping, into a :vimgrep pattern that is run through :execute. Because :vimgrep honors the bar as a command separator, a crafted tag field can close the search pattern and append an arbitrary Ex command; opening a hostile .c file whose project tags file contains such an entry and invoking C omni-completion runs that command as the editing user. This issue is fixed in version 9.2.0735.Show less

Vim is an open source, command line text editor. Prior to 9.2.0736, the PHP omni-completion script in runtime/autoload/phpcomplete.vim interpolates a class or trait name, taken from the contents of the edited buffer, into a search() pattern that is run via win_execute() without escaping. A name containing a single quote can terminate the search() string argument early, and because the bar is honored as an Ex command separator, the remainder of the name is run as Ex commands; via the :! command this allows arbitrary operating-system command execution when a victim opens a crafted PHP file and invokes omni-completion. This issue is fixed in version 9.2.0736.Show less

SiYuan is an open-source personal knowledge management system. Prior to 3.7.1, SiYuan renders note and package content to HTML through the Lute engine with sanitization enabled, but Lute's dangerous javascript scheme block does not check form action or SVG xlink:href attributes, allowing stored cross-site scripting in document export-preview and Bazaar package README render paths that can execute OS commands in the Electron desktop renderer. This issue is fixed in versions 3.7.1.Show less

Metabase is an open-source business intelligence and embedded analytics tool. From 1.55.0 until 1.58.15.1, 1.59.12, 1.60.6.3, and 1.61.2, Metabase did not validate unsafe H2 connection properties on one database-creation code path, allowing an authenticated administrator to register a crafted H2 database connection and execute arbitrary Java code on the Metabase server. This issue is fixed in versions 1.58.15.1, 1.59.12, 1.60.6.3, and 1.61.2.Show less

A security vulnerability has been detected in YzmCMS up to 7.5. Affected is the function get_url of the file /yzmphp/yzmphp.php of the component Header Handler. The manipulation of the argument HTTP_HOST leads to cross site scripting. The attack may be initiated remotely. The exploit has been disclosed publicly and may be used. The vendor was contacted early about this disclosure but did not respond in any way.Show less

A weakness has been identified in apidevtools json-schema-ref-parser up to 15.3.5. This impacts the function Refs.set/Pointer.set in the library lib/pointer.ts. Executing a manipulation can lead to improperly controlled modification of object prototype attributes. The attack can be launched remotely. Upgrading to version 15.3.6 will fix this issue. This patch is called a786bc6afc3674f650496472ee93d5cf74c4bd84. It is suggested to upgrade the affected component.Show less

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. Prior to 0.10.0, get_event_call delivered execute:python and execute:tool Socket.IO events to a client-supplied session_id after checking only that the session was connected, allowing authenticated users who learned another socket ID through ydoc:document:join to run code interpreter Python or tools in that user session. This issue is fixed in version 0.10.0.Show less

A security flaw has been discovered in enquirer up to 2.4.1. Affected is the function Enquirer.set of the component Public Package API. The manipulation of the argument question.name results in improperly controlled modification of object prototype attributes. The attack can be launched remotely. The exploit has been released to the public and may be used for attacks. The project was informed of the problem early through an issue report.Show less

OpenCTI is an open source platform for managing cyber threat intelligence knowledge and observables. Prior to 7.260401.0, the OpenCTI GraphQL API exposes a script filter operator in its FilterOperator enum that allows any authenticated user with the KNOWLEDGE capability to pass user-supplied Elasticsearch Painless script values directly into search queries without validation or sanitization, allowing computationally expensive scripts to consume cluster CPU resources and degrade or deny service for all users. This issue is fixed in version 7.260401.0.Show less

LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. Prior to 1.82.0-stable, LiteLLM's Custom Code Guardrails production create and update paths did not apply the same sandboxing and validation used by the test endpoint, allowing a privileged user with access to create or update guardrails to submit custom Python code that executed in the LiteLLM proxy environment and could expose secrets available to the process. This issue is fixed in version 1.82.0-stable.Show less

Copier is a library and CLI app for rendering project templates. In versions 9.5.0 through 9.15.1, the `trust` setting's prefix match (`copier/_settings.py`) compares the template URL against a trusted prefix with a raw `str.startswith` and no path normalization, while the URL is normalized when the template is actually fetched (`Path.resolve()` for local paths; libcurl dot-segment removal for `https`). A template reference that textually starts with a trusted prefix but contains `..` is therefore granted trust yet resolves to a different, attacker-controlled template, whose `tasks` / `migrations` / `jinja_extensions` then run without the `--trust` prompt — arbitrary command execution. Version 9.15.2 patches the issue.Show less

Koodo Reader is an ebook reader. In version 2.3.0 and earlier, Koodo Reader is vulnerable to remote code execution through malicious EPUB files because the open-book IPC handler enables nodeIntegrationInSubFrames and EPUB chapter content is rendered with unsanitized innerHTML. An attacker can craft an EPUB book that, when imported and opened by the victim, instantiates a hidden iframe with Node.js API access and executes arbitrary operating system commands with the victim user's privileges. This issue is fixed in version 2.3.1.Show less