The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains a dynamic context execution block (!command) designed to run at skill load time. This command executes shell logic to check for and run a Table of Contents utility (skill-toc) from either the system PATH or a specific local directory ($HOME/.claude/skills/plugin-utils/bin/skill-toc). While intended for documentation generation, this allows for arbitrary shell execution upon opening the skill.
[COMMAND_EXECUTION]: The skill instructions and reference files contain multiple instances of shell command execution. This includes the use of xan for CSV manipulation and curl for API interactions. These commands are intended to be executed by the agent to perform data analysis and transport verification.
[INDIRECT_PROMPT_INJECTION]: The skill is specifically designed to ingest and process data from third-party websites (e.g., SEC EDGAR, FINRA BrokerCheck) and professional datasets. This creates a surface for indirect prompt injection, as the agent may process instructions hidden within the scraped content.
Ingestion points: External data fetched from scraping zones or Bright Data's Web Archive, as well as CSV manifests processed in SKILL.md.
Boundary markers: The instructions focus on technical constraints such as rate limits and quotas but lack specific guidance or delimiters to prevent the agent from following instructions embedded in the external data.
Capability inventory: The skill possesses capabilities for network requests (urllib.request), file system access (reading credentials from ~/.config/brightdata/token), and shell command execution (xan, curl).
Sanitization: The provided Python scripts and instructions do not demonstrate sanitization or validation of the retrieved content to mitigate against malicious instruction injection.

web-scraping