The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes a Python script (scripts/find_code_language.py) that fetches documentation from a well-known external source.
Evidence: The script connects to https://raw.githubusercontent.com/highlightjs/highlight.js/main/SUPPORTED_LANGUAGES.md to retrieve the latest list of supported languages and aliases.
Context: Highlight.js is a standard industry library for syntax highlighting; fetching its documentation is a legitimate operation for the skill's purpose.
[COMMAND_EXECUTION]: The skill uses local command execution for utility and testing purposes.
Evidence: The scripts/test_find_code_language.py file uses subprocess.run to execute the local Python helper script.
Context: This execution is limited to the skill's own internal logic and does not involve executing untrusted strings or external payloads.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from a remote repository, which presents a theoretical attack surface for indirect prompt injection.
Ingestion points: scripts/find_code_language.py fetches Markdown content from an external GitHub URL.
Boundary markers: None are explicitly used for the external data, but the parsing logic is strictly limited to Markdown table structures via regular expressions.
Capability inventory: The skill can generate Markdown and execute its bundled Python helper script.
Sanitization: The script parses the remote Markdown into a structured JSON payload. This structured data approach acts as an effective sanitization layer, preventing raw text from the remote source from being interpreted as direct instructions by the agent.

azure-devops-wiki-markdown