The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill operates by having the agent copy TypeScript templates to the /tmp directory, adapt them by inserting user-provided content, and then execute the resulting script using npx tsx. This runtime generation and execution of code is a significant capability that requires careful oversight.
[INDIRECT_PROMPT_INJECTION]: The process of interpolating user data into template-based scripts lacks explicit sanitization or validation logic, creating a surface for potential code injection within the generated file.
Ingestion points: Document content, titles, and data values provided by the user to the agent.
Boundary markers: None; the instructions do not define specific delimiters to separate untrusted user data from the script logic during template adaptation.
Capability inventory: The environment allows writing files to the filesystem and executing shell commands via npx tsx.
Sanitization: There are no instructions for the agent to sanitize, escape, or validate user input before placing it into the script templates in agents/office-docs.md.
[PROMPT_INJECTION]: The agent instructions in agents/office-docs.md explicitly direct the agent to 'execute silently' and 'don't show the code' to the user. While intended for token efficiency, this concealment reduces transparency and prevents user review of the dynamically generated code before execution.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands to install dependencies (npm install) and execute the generated document scripts (npx tsx).
[METADATA_POISONING]: There is a discrepancy between the author identified in the skill context ('ma1orek') and the author listed in the plugin metadata ('Jeremy Dawes' in .claude-plugin/plugin.json). While not necessarily malicious, such inconsistencies can be used to misrepresent the origin and trustworthiness of a skill.

office