The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's 'Build mode' workflow requires the agent to copy a script template (scripts/tor_docx.py), programmatically populate it with user-provided procurement details, and then execute the resulting Python file. This runtime generation and execution of code based on user-supplied parameters introduces a potential injection vector where malicious user input could be interpreted as code.
[REMOTE_CODE_EXECUTION]: The agent is instructed to use its environment to run the generated Python scripts to render .docx deliverables. While the template itself is local, the execution of dynamically generated code is a sensitive capability.
[PROMPT_INJECTION]: The skill is exposed to indirect prompt injection in 'Review' and 'Finalise' modes when processing existing documents.
Ingestion points: The agent reads external content from Markdown files, Python scripts, and .docx files (using python-docx) to perform reviews and edits.
Boundary markers: There are no specific instructions or delimiters defined to separate untrusted document content from the agent's core instructions, nor are there warnings to ignore embedded commands.
Capability inventory: The agent has the ability to write to the filesystem (via the Edit tool in 'Finalise' mode) and execute code ('Build' mode).
Sanitization: The skill does not define any sanitization, validation, or escaping procedures for the text extracted from external procurement documents.

tor-procurement