The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from a user-specified Obsidian vault, including PDFs, transcripts, and Markdown notes. This creates a surface for indirect prompt injection, where malicious instructions embedded in these external documents could potentially influence the agent's behavior during synthesis or research tasks.
Ingestion points: The agent reads from the absolute vault path configured in .agent/context/knowledge-vault.md and processes files within that directory (e.g., SKILL.md, knowledge-vault-template.md).
Boundary markers: While the skill defines conventions for note structure (Title Case, wikilinks), it does not specify explicit delimiters or "ignore instructions" prompts to isolate the content of ingested research materials.
Capability inventory: The skill utilizes filesystem tools (Read, Write, Glob, Grep) and suggests the use of shell commands like pdftotext, find, and grep for data extraction and search (SKILL.md, knowledge-vault-template.md).
Sanitization: There is no evidence of sanitization or filtering of external content before it is processed by the model.
[COMMAND_EXECUTION]: The skill instructions and recipes suggest that the agent should execute shell commands for maintenance and data extraction tasks.
Evidence: SKILL.md and knowledge-vault-template.md provide example commands such as `pdftotext "/path/to/Paper.pdf"
| head -200andgrep -rl "[[Some Concept]]" "{{VAULT_PATH}}"`. While these are routine for vault management, they represent an active use of the system's command-line interface.

setup-knowledge-vault