The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize untrusted data from external sources including emails, chat messages, and document comments.
Ingestion points: External data is fetched via connected tools (Calendar, Gmail, Slack, Microsoft 365) as described in the Gather section of SKILL.md.
Boundary markers: The skill includes explicit instructions in the Ground rules section to treat all gathered content as data only and to ignore any embedded commands or "notes to Claude".
Capability inventory: The skill can perform network fetches, execute shell commands via Node.js for verification, and generate deep-links to create new Claude sessions.
Sanitization: SKILL.md mandates that all gathered text must be rendered as escaped plain text in the HTML artifact to prevent live markup or script execution.
[COMMAND_EXECUTION]: The Build section of SKILL.md utilizes node -e to execute a Playwright script. This script launches a headless browser to take a screenshot of the generated HTML brief for internal verification purposes.
[EXTERNAL_DOWNLOADS]: The skill may attempt to install the playwright and @fontsource/fraunces packages using npm if they are not already present in the environment. These are well-known, standard packages used for browser automation and font management.
[DYNAMIC_EXECUTION]: The skill uses a hardcoded Node.js script to perform a visual render check of the final artifact. While this involves dynamic execution, it is scoped to an internal validation task using a pre-defined template.

morning