The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches module metadata from the official Nextflow Registry (registry.nextflow.io). Since this is the primary vendor repository for the Nextflow ecosystem, these external references are expected and safe within the tool's context.
[REMOTE_CODE_EXECUTION]: A curl command piped to python3 is used to parse JSON data from the registry API. While this pattern was flagged by automated scanners, technical analysis confirms the Python script only extracts and prints metadata strings for display and does not execute instructions from the downloaded content.
[COMMAND_EXECUTION]: The skill utilizes standard bioinformatics tools and Nextflow commands (e.g., nextflow module search, nextflow module run) to manage workflows. These operations are consistent with the skill's stated purpose and use the agent's provided Bash tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data, creating a potential surface for indirect injection.
Ingestion points: Metadata (names and descriptions) retrieved from the registry.nextflow.io API in Step 1.
Boundary markers: Not explicitly defined in the raw API call, though the overall process separates the planning and execution phases.
Capability inventory: The skill has access to Bash, Write, and can delegate execution to Skill(skill="run-module").
Sanitization: External data is printed to the agent context without specific sanitization; however, a mandatory human-in-the-loop checkpoint ('Step 2: User Agreement') requires explicit user review and approval of the proposed plan before any capabilities are invoked.

create-workflow