The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for the agent to install a tool by fetching scripts from a third-party GitHub repository and piping them directly to shell interpreters. This method, using curl | bash for Linux/macOS and iwr | iex for Windows, allows for immediate execution of remote code without manual verification.
Evidence: curl -sSL https://raw.githubusercontent.com/asynkron/Asynkron.QuickDup/main/install.sh | bash found in SKILL.md and references/quickdup.md.
Evidence: iwr -useb https://raw.githubusercontent.com/asynkron/Asynkron.QuickDup/main/install.ps1 | iex found in SKILL.md and references/quickdup.md.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the repository being scanned and uses that data to influence its subsequent actions.
Ingestion points: The skill reads file contents (.cs, .razor) and local configuration files (.quickdup/results.json, .quickdup/ignore.json) during the discovery and refactoring workflow.
Boundary markers: There are no explicit instructions or delimiting markers to ensure the agent ignores potentially malicious instructions embedded within the scanned files.
Capability inventory: The skill possesses the ability to modify files and execute repository-specific build and test commands, which could be abused if an injection attack succeeds.
Sanitization: No sanitization or validation mechanisms are described for the content extracted from external files before it is processed by the agent.

quickdup