The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches report source data from a public GitHub repository via the UpdateSources.ts tool. It also downloads report content from various cybersecurity vendor websites using the FetchReport.ts tool to facilitate summarization and analysis.
[COMMAND_EXECUTION]: The skill executes internal logic using Bun-based TypeScript tools for fetching, listing, and updating its data sources.
[PROMPT_INJECTION]: The SKILL.md file contains instructions for the agent to load and apply instructions from a customization directory (~/.claude/skills/CORE/USER/SKILLCUSTOMIZATIONS/AnnualReports/), which allows external files to override default behavior.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it fetches and processes untrusted external report content.
Ingestion points: External report content fetched by Tools/FetchReport.ts and the upstream source list fetched by Tools/UpdateSources.ts.
Boundary markers: No explicit delimiters or instructions are used to prevent the agent from following commands embedded in fetched reports.
Capability inventory: The skill has network access (fetch), file system access (fs), and tool execution capabilities (bun run) across Tools/FetchReport.ts, Tools/UpdateSources.ts, and Tools/ListSources.ts.
Sanitization: The Tools/FetchReport.ts tool performs basic HTML tag stripping but lacks mechanisms to sanitize or validate the text content for potential malicious instructions.

AnnualReports