The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes explicit instructions that override default agent behavior and response styles. Evidence: 'Rule 1: NEVER Ask Clarifying Questions', 'You have commands to answer your own questions. USE THEM.', and 'Rule 2: Always End With a Text Summary. NEVER end your turn with only a command execution' in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external CLI tool from the NuGet registry. Evidence: 'dotnet tool install --global Sbroenne.PowerPointMcp.CLI' in SKILL.md.
[COMMAND_EXECUTION]: The skill relies on and manages a local background service and CLI tool to perform operations. Evidence: 'pptcli service start', 'pptcli service stop --force', and various 'pptcli' command groups described in references/cli-commands.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input that is interpolated into commands and slide content, creating a potential vector for malicious data to influence the agent.
Ingestion points: Untrusted data enters the agent context via file paths (filePath in SKILL.md), image/media paths (image_path in images.md, media_path in media.md), and slide text content (text in text-formatting.md).
Boundary markers: The instructions lack explicit boundary markers or 'ignore embedded instructions' prompts to distinguish user-provided content from agent commands.
Capability inventory: The skill utilizes 'pptcli' to execute shell commands that interact with the Windows host via COM interop, allowing for file system reads/writes (pptx, pdf, images) and process management (pptcli service stop).
Sanitization: There is no mention of sanitization, escaping, or validation of user-supplied text or paths before they are used in CLI commands or slide content.

powerpoint-cli