The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads several external tools and data sources during its preparation and execution phases:
Prepare.bat and prepare.sh fetch the uv Python package manager from https://astral.sh and the busybox.exe utility from https://frippery.org.
download_pluginhub.py and download_magnetarhub.py clone plugin registries from https://github.com/StarCpt/PluginHub.git and https://github.com/CometWorks/magnetar-hub.git.
download_plugin_source.py clones arbitrary plugin source code repositories from GitHub based on the contents of the retrieved registries.
[REMOTE_CODE_EXECUTION]: The preparation scripts utilize remote execution patterns to install system dependencies:
Prepare.bat executes a remote PowerShell script for uv installation via irm https://astral.sh/uv/install.ps1 | iex.
prepare.sh and common-posix.sh pipe a remote shell script for uv installation to sh using curl -LsSf https://astral.sh/uv/install.sh | sh.
[COMMAND_EXECUTION]: Multiple Python and shell scripts execute system commands to manage the environment and source code:
download_pluginhub.py, download_magnetarhub.py, and download_plugin_source.py use subprocess.run to invoke the git CLI for repository cloning, fetching, and checkout operations.
download_plugin_source.py triggers the local index_plugins.py script using uv run.
common-posix.sh uses dotnet tool install to install the ilspycmd utility.
[PRIVILEGE_ESCALATION]: Prepare.bat invokes PowerShell with the -ExecutionPolicy ByPass flag to run the remote installation script for the uv tool, overriding local execution restrictions for that operation.
[INDIRECT_PROMPT_INJECTION]: The skill clones, indexes, and searches external C# source code from GitHub repositories. This behavior creates an attack surface where malicious instructions or deceptive metadata embedded in the external code (e.g., in comments or string literals) could be ingested into the agent's context during searches.
Ingestion points: External C# files cloned into Data/Sources are parsed and indexed by index_plugins.py.
Boundary markers: The search results produced by search_plugins.py do not implement explicit delimiters or warnings to ignore embedded instructions in the retrieved code snippets.
Capability inventory: The skill has capabilities for network access (requests, git), file system modification (mkdir, mklink, write), and subprocess execution (git, powershell, uv).
Sanitization: There is no evidence of sanitization or filtering of the external source code content before it is indexed or returned to the agent context.

se-dev-plugin