The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions in SKILL.md (Step 4) direct the agent to execute a shell command curl -sL "<url>" where the URL is dynamic and sourced from external search results. A malicious search result could provide a URL containing shell metacharacters (e.g., semicolons, backticks, or pipes), leading to arbitrary command execution on the host system when interpolated into the command string.
[REMOTE_CODE_EXECUTION]: The workflow involves fetching raw data from external, untrusted web sources and piping it directly into a Python interpreter (python3 -c) for processing. While the provided inline Python script is intended for text extraction, piping unvalidated remote content into an interpreter environment creates a significant security risk if the input is manipulated.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) due to its core functionality of processing external data.
Ingestion points: The skill fetches full-page text content from arbitrary URLs found via DuckDuckGo (Step 4 in SKILL.md).
Boundary markers: Absent. The instructions do not provide delimiters or warnings to the agent to treat the fetched content as untrusted or to ignore instructions embedded within that content.
Capability inventory: The skill has the ability to execute shell commands, perform network requests via curl, and write files to the filesystem (mkdir and report saving in Step 6).
Sanitization: The skill performs only basic regex-based HTML tag removal; it lacks any mechanism to sanitize the semantic content of the fetched text to prevent it from influencing the agent's logic or synthesis process.

deep-research-pro