The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface due to its data processing pipeline.
Ingestion points: scripts/crawl.py fetches untrusted external web content and stores it in evidence/pages.jsonl for subsequent analysis.
Boundary markers: Untrusted web content is interpolated into LLM prompts in scripts/bootstrap.py (for brand fact extraction) and scripts/sample.py (for AI answer sampling) without explicit delimiters or instructions to ignore embedded directives.
Capability inventory: The skill possesses significant capabilities including local command execution (scripts/jobs.py), external network publishing to GitHub, WordPress, and WeChat (scripts/publish.py), and file system write operations.
Sanitization: Content is stripped of HTML boilerplate but is not sanitized or filtered for potential prompt injection markers before being passed to LLM providers.
[COMMAND_EXECUTION]: The skill executes local shell commands through a dashboard-driven job system.
The scripts/jobs.py module uses subprocess.Popen to run background tasks. Security is maintained by using a list-based argument format and limiting execution to a strict whitelist of subcommands and parameters defined in the ACTIONS dictionary.
User-provided project identifiers (slugs) are validated against a strict alphanumeric regex in scripts/geolib.py before being used in file paths or command lines.
[DATA_EXFILTRATION]: The skill transmits project data and credentials to external services as part of its core functionality.
Brand facts and diagnostic questions are sent to LLM providers including 302.AI, OpenRouter, and native endpoints (OpenAI, Anthropic, etc.).
The publication system in scripts/publish.py transmits generated content to external platforms using credentials managed via the environment. Access to these operations is gated by a local-only dashboard (127.0.0.1) and manual user triggers.
[EXTERNAL_DOWNLOADS]: The skill performs automated web crawling of user-specified target domains.
scripts/crawl.py implements a crawler that fetches HTML content. It includes safety guards such as a 4MB per-page download limit and a file extension blacklist to prevent the ingestion of malicious binaries or excessively large files.

geo