The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SKILL.md file contains a command that downloads a shell script from a remote URL and pipes it to the shell: curl -fsSL https://raw.githubusercontent.com/colbymchenry/codegraph/main/install.sh | sh. This practice allows for arbitrary code execution on the user's system without prior inspection of the remote content.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install software via npx @colbymchenry/codegraph and npm i -g @colbymchenry/codegraph. The target package is maintained by an external entity that does not align with the skill author context ('openlark'), posing a risk of executing untrusted third-party code.
[COMMAND_EXECUTION]: The skill defines various CLI commands, such as codegraph install and codegraph serve, which involve executing binary files and modifying system configurations. These commands are executed locally and could be leveraged for malicious purposes if the downloaded software is compromised.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection.
Ingestion points: External source code is processed and stored during codegraph index operations.
Boundary markers: There are no documented delimiters to separate indexed data from agent instructions.
Capability inventory: The agent is granted access to tools like codegraph_explore which expose code contents.
Sanitization: There is no evidence of content sanitization to prevent the agent from following instructions embedded within the code being analyzed.

codegraph-tool