The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted external sources (web research) and local user files (Obsidian vault transcripts) to dynamically generate new prompt files. Malicious content within these sources could potentially influence the behavior of agents that subsequently run the generated prompts.
Ingestion points: Uses WebSearch and WebFetch for company research and Grep to extract text from transcripts and contacts in the Obsidian vault.
Boundary markers: The generated prompt template (references/prompt-template.md) lacks boundary markers, delimiters, or warnings to ignore embedded instructions within the interpolated data.
Capability inventory: The skill possesses the ability to write files to the local filesystem and transmit data through an external Telegram tool.
Sanitization: No sanitization or validation of the research or transcript text is performed before it is inserted into the generated prompt template.
[DATA_EXFILTRATION]: The skill is designed to search through sensitive local data (call transcripts and contact lists) and provides a specific workflow to transmit the compiled information to an external service.
Sensitive data access: Explicitly targets the People/ folder and Daily/ transcript logs in the user's Obsidian vault.
External transmission: Features a user-triggered Save and send via Telegram option which uses an external tool to move generated documents out of the local environment.

context-builder