The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides explicit instructions to download a PowerShell installation script from a remote URL (https://raw.githubusercontent.com/DeusData/codebase-memory-mcp/main/install.ps1) and execute it locally. This pattern of fetching and running remote scripts from untrusted sources is a high-risk activity.
[EXTERNAL_DOWNLOADS]: References external resources for installation and setup, specifically from the DeusData GitHub repository, which is not an established trusted vendor. It also references a proxy script cbm-dashboard-plus.py located in a local tools directory.
[COMMAND_EXECUTION]: The skill utilizes PowerShell command execution (&) to run local binaries and scripts. It also uses Invoke-RestMethod to interact with a local RPC service on port 9749 for querying project status and managing graph data.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The index_repository tool ingests the entire content of a codebase, including comments and file names that could contain adversarial instructions.
Boundary markers: No specific boundary markers or instructions to ignore embedded prompts are mentioned when the agent processes content retrieved via search_code or get_code_snippet.
Capability inventory: The skill allows the agent to read source code, execute Cypher queries on the graph, and delete indexed projects, which could be abused if the agent is manipulated by injected instructions in the source code.
Sanitization: There is no evidence of sanitization or safety filtering for the data indexed into the knowledge graph.

codebase-memory