The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill triggers the download and installation of system-level dependencies. According to references/export-guide.md, the export-pdf.sh script installs Playwright and Chromium (~150MB), and deploy.sh checks for or installs the Vercel CLI. Additionally, references/pptx-conversion.md requires installing Python packages via pip. These are established tools for their respective tasks.
[COMMAND_EXECUTION]: The skill utilizes shell scripts for its core export and deployment functionality.
Evidence: SKILL.md and references/export-guide.md instruct the use of bash "${CLAUDE_PLUGIN_ROOT:-.}"/scripts/export-pdf.sh and bash "${CLAUDE_PLUGIN_ROOT:-.}"/scripts/deploy.sh to handle artifact processing.
[DATA_EXFILTRATION]: There is an inherent risk of data exposure due to the skill's design, which processes sensitive corporate data for public presentation.
Evidence: references/task-method.md directs the agent to extract goals and outcomes from "CRM opportunity/account" and "renewal calls".
Evidence: references/export-guide.md describes a "Live URL (Vercel Deploy)" feature for sharing decks as public links, which could lead to sensitive internal metrics being exposed if not managed carefully.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface where malicious instructions in researched data could influence the agent.
Ingestion points: The agent reads external CRM data and prior call records as described in references/task-method.md.
Boundary markers: None are specified to protect the agent from instructions embedded in that research data.
Capability inventory: The skill can execute shell scripts, write files via Python, and deploy content to the web.
Sanitization: No sanitization or escaping protocols are mentioned for the interpolated research data before it is incorporated into the presentation.
[DYNAMIC_EXECUTION]: The skill uses Python scripts to perform file operations during PPTX conversion.
Evidence: references/pptx-conversion.md contains code that uses with open(fname, "wb") as f: f.write(img.blob) to dynamically save extracted images to the filesystem.

deck