The Agent Skills Directory

[DYNAMIC_EXECUTION]: Runtime compilation and injection logic. The file scripts/office/soffice.py dynamically generates a C source file and invokes gcc to compile it into a shared object library at runtime. This library is then injected into the soffice process environment via the LD_PRELOAD environment variable to override network-related system calls.
[COMMAND_EXECUTION]: Execution of system binaries. Multiple scripts including scripts/recalc.py, scripts/office/soffice.py, and scripts/office/validators/redlining.py invoke external tools such as gcc, soffice, git, and timeout using subprocess.run with parameters determined by the skill's logic.
[INDIRECT_PROMPT_INJECTION]: Processing of untrusted user data. The skill is designed to ingest and process user-provided spreadsheet files which represent a potential injection surface:
Ingestion points: User-supplied spreadsheet files (.xlsx, .csv, etc.) enter the agent context as described in SKILL.md.
Boundary markers: Absent. No specific delimiters or warnings are used to isolate external content.
Capability inventory: The skill has significant capabilities including arbitrary command execution and file system access.
Sanitization: Partial sanitization is provided through the use of defusedxml in several validation modules.

xlsx