The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions for both Linux/macOS and Windows utilize piped remote execution (curl | sh and irm | iex) targeting scripts hosted on GitHub (raw.githubusercontent.com/JheisonMB/). This pattern is a significant security risk as it allows for arbitrary code execution without prior review of the script content.
[REMOTE_CODE_EXECUTION]: The skill documentation indicates that the texforge build command automatically downloads and executes the tectonic binary from the internet during its first run. Dynamic execution of remote binaries bypasses static analysis and standard package management safety checks.
[EXTERNAL_DOWNLOADS]: The texforge template add command fetches LaTeX templates from a remote registry, introducing external content into the local environment.
[COMMAND_EXECUTION]: The skill provides numerous commands (new, build, fmt, check) that perform filesystem operations and execute CLI logic based on user-provided project files.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface. It processes untrusted LaTeX and bibliography files (.tex, .bib) which are ingested into the tool's context. There are no specified boundary markers or sanitization steps to prevent embedded instructions in these files from influencing the agent's behavior, particularly given the tool's capability to perform network operations and binary execution.
Ingestion points: Reads .tex and .bib files (SKILL.md)
Boundary markers: None present
Capability inventory: Subprocess calls for compilation, file-write to build/, network operations for templates (SKILL.md)
Sanitization: None described

texforge