The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is designed to scrape documentation content from official sources at docs.duendesoftware.com. Specifically, it targets the llms-full.txt endpoint to maintain a local canonical store. This is the primary function of the skill and uses a trusted vendor domain.
[COMMAND_EXECUTION]: The skill utilizes Python scripts and the Bash tool to perform maintenance tasks. duende_docs_api.py executes internal scripts like refresh_index.py using subprocess.run. These executions are limited to the skill's own directory and are used for index management.
[PROMPT_INJECTION]: SKILL.md contains strict operational instructions intended to override default AI agent tendencies that could lead to technical failure. These include constraints against using cd && in PowerShell to avoid path doubling and a prohibition on using the read_file tool on the large index.yaml file to prevent context-limit errors. These are legitimate reliability guardrails rather than malicious injections.
[PROMPT_INJECTION]: An indirect prompt injection surface exists as the skill processes external documentation content.
Ingestion points: scripts/core/llms_full_parser.py and scripts/core/scrape_docs.py (implied) ingest content from external URLs into the agent's environment.
Boundary markers: The parser uses ----- separators, but the instructions do not explicitly warn the agent to ignore instructions embedded within the documentation text.
Capability inventory: The skill has access to command execution via Python and Bash tools.
Sanitization: Content is parsed for metadata, but there is no evidence of sanitization to filter out natural language instructions targeting the AI agent.

duende-docs