The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads from a hardcoded sensitive file path within the user's local directory: /Users/mtm/Documents/Obsidian Vault/apples and plums near us.md. This note is treated as a source of truth for location data and is processed by the agent.
[DYNAMIC_EXECUTION]: The instructions explicitly direct the agent to generate a "short script" at runtime, write it to a temporary directory ($CLAUDE_JOB_DIR/tmp), and execute it using uv run --no-active. This bypasses static analysis and introduces a risk of arbitrary code execution if user-supplied data or vault content is improperly handled during code generation.
[COMMAND_EXECUTION]: The skill makes extensive use of the curl command to interact with external services and resolve URLs.
It uses curl --location to expand user-provided short links, which could be exploited for Server-Side Request Forgery (SSRF) or command injection if the input is not sanitized.
It uses uv run to execute generated Python scripts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external source (the Obsidian vault note).
Ingestion points: The note file /Users/mtm/Documents/Obsidian Vault/apples and plums near us.md is read to extract coordinates and headings.
Boundary markers: None identified. The skill does not provide instructions to ignore potentially malicious content embedded in the note.
Capability inventory: The skill has the ability to execute shell commands (curl), generate and run Python scripts (uv run), and perform network operations.
Sanitization: The skill lacks robust sanitization for content read from the vault note before it is interpolated into shell commands or generated scripts.
[EXTERNAL_DOWNLOADS]: The skill sends location data and vault content to external services including Nominatim (OpenStreetMap), Overpass API, and OSRM (Open Source Routing Machine). These are well-known services for geographic data processing.

foraging-route