The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions promote a dangerous execution pattern by downloading a remote shell script and piping it directly to bash.
Evidence: curl -fsSL https://raw.githubusercontent.com/yvgude/lean-ctx/main/skills/lean-ctx/scripts/install.sh | bash in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill fetches resources and installation scripts from a personal GitHub repository that is not associated with a recognized trusted organization.
Source URL: https://raw.githubusercontent.com/yvgude/lean-ctx/main/.
[COMMAND_EXECUTION]: The tool's primary purpose involves wrapping and executing arbitrary shell commands through its CLI.
The pattern lean-ctx -c <command> allows the agent to execute any command in the shell environment to capture and compress output.
[PERSISTENCE_MECHANISMS]: The setup process involves installing shell hooks and global aliases, which modifies the user's shell profile to persist tool functionality across sessions.
Evidence: lean-ctx init --global and lean-ctx setup instructions.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting and processing data from untrusted files and command outputs.
Ingestion points: Data enters the context via lean-ctx read <file> and the output of commands executed through lean-ctx -c.
Boundary markers: The skill documentation does not specify the use of clear delimiters or instructions to ignore embedded commands in the compressed output.
Capability inventory: The skill has the ability to read arbitrary files and execute shell commands.
Sanitization: No sanitization or filtering of the ingested content is mentioned.

lean-ctx