The Agent Skills Directory

[COMMAND_EXECUTION]: The agent is instructed to use mkdir -p /tmp/tear-sheet/ to create a temporary working directory for intermediate data files during the generation process.
[DYNAMIC_EXECUTION]: The skill provides template JavaScript functions and instructs the agent to generate a Node.js script to assemble the final DOCX document. This follows a low-risk pattern of generating code from provided, static templates.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data, such as earnings call transcripts and company summaries from S&P Global tools.
Ingestion points: Data is retrieved via S&P Global MCP tools and written to intermediate files like /tmp/tear-sheet/earnings.txt and /tmp/tear-sheet/company-profile.txt in SKILL.md Step 3.
Boundary markers: The instructions do not define explicit structural delimiters but provide clear intent for content isolation.
Capability inventory: The skill has capabilities to create directories, write local files, and execute Node.js code for document generation as detailed in SKILL.md Step 3 and Step 4.
Sanitization: Content Quality Rule 11 in SKILL.md explicitly mandates that the agent must rewrite every narrative section (e.g., company summaries, earnings highlights) for the specific audience rather than pasting external content verbatim, which serves as a primary mitigation against embedded malicious instructions.

tear-sheet