The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md file contains instructions for the agent to execute a shell command python3 scripts/install-external-skills.py with specific arguments to extend the agent's functionality at runtime.
[REMOTE_CODE_EXECUTION]: The skill implements a workflow for installing "external skills" based on a registry file (external-skills.yaml). This pattern facilitates the retrieval and execution of external code that is not part of the analyzed skill package, which can lead to the execution of unverified logic.
[INDIRECT_PROMPT_INJECTION]: The skill's primary focus is "Context Engineering," which involves a "Context Assembly Flow" that specifically includes preprocessing uploaded files and scraping URLs. This ingestion of untrusted external content into the agent's prompt context represents a significant attack surface for indirect prompt injection.
Ingestion points: ContextAssemblyService.preprocess_inputs (files, links), URL scraping functions, and external API responses via MCP (Model Context Protocol).
Boundary markers: The documentation mentions the use of XML-style system_block tags for System Prompt Learning (SPL) blocks.
Capability inventory: The skill documentation references tools such as create_file, update_profile, update_job_posting, and various vector store query capabilities.
Sanitization: The skill mentions that BDI (Beliefs-Desires-Intentions) states are HTML-escaped and length-limited, but does not provide comprehensive sanitization details for scraped web content.
[DYNAMIC_EXECUTION]: The "SPL Processor" and build_dynamic_instructions() mechanisms described in the memory documentation allow for the dynamic assembly and injection of instructions into the system prompt at runtime, based on session variables and external signals.

context-engineering