The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides installation commands that download shell scripts from a remote server and pipe them directly into the bash interpreter (curl | bash). This bypasses local inspection and allows for arbitrary code execution if the remote repository is compromised.
Evidence: curl -fsSL https://raw.githubusercontent.com/DeusData/codebase-memory-mcp/main/install.sh | bash (SKILL.md)
Evidence: curl -fsSL https://raw.githubusercontent.com/DeusData/codebase-memory-mcp/main/install.sh | bash -s -- --ui (SKILL.md)
[REMOTE_CODE_EXECUTION]: The skill provides PowerShell commands for Windows that download and execute a script file, posing a similar risk to the piped bash execution.
Evidence: Invoke-WebRequest -Uri https://raw.githubusercontent.com/DeusData/codebase-memory-mcp/main/install.ps1 -OutFile install.ps1; .\install.ps1 (SKILL.md)
[EXTERNAL_DOWNLOADS]: The skill relies on binaries and scripts hosted on GitHub under the 'DeusData' account, which is not a verified or trusted organization in the security context. While GitHub is a well-known service, the specific account/vendor lacks established trust for automatic execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and index large codebases (up to 28M LOC). This presents a massive attack surface for indirect prompt injection, as malicious instructions hidden in comments, strings, or documentation within the codebase could be interpreted as commands by the agent during analysis.
[COMMAND_EXECUTION]: The skill documentation includes several CLI commands (codebase-memory-mcp cli ...) that are intended to be run locally. While these are for tool operation, the automated installation method makes the subsequent execution of these commands potentially unsafe if the initial binary was malicious.

codebase-memory-mcp-intelligence