The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands (grep, curl, defuddle, git, rm) using user-provided URLs. Specifically, URLs are placed inside double quotes in commands like defuddle parse "<URL>" or used in curl requests. If a URL contains shell metacharacters such as double quotes or backticks, it could lead to arbitrary command execution on the host system since no explicit escaping or sanitization is described before command construction.
[EXTERNAL_DOWNLOADS]: The skill uses curl and the defuddle tool to fetch content and metadata from external URLs provided by the user. While core to its function, this involves interaction with untrusted remote resources.
[DATA_EXFILTRATION]: In Step 2, the skill uses grep to read all markdown files (*.md) in the current directory to check for duplicates. Although used for vault organization, this grants the skill the ability to read and potentially expose content from the local filesystem.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection. It fetches titles and summaries from external websites and presents them in a preview table (Step 3) or processes them via subagents (Steps 4 and 5.5). Ingestion points: User-provided URLs and fetched web metadata. Boundary markers: URLs are quoted in prompts but lack sanitization. Capability inventory: Filesystem access, network requests, and subagent tasking. Sanitization: None described. Malicious content in a webpage title could influence the agent's behavior during the review or save phases.

batch-save