The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input through the $ARGUMENTS variable to define the 'idea' for an experiment. This creates a surface for indirect prompt injection if the user provides malicious instructions instead of a project idea.
Ingestion points: SKILL.md (via $ARGUMENTS variable).
Boundary markers: No specific delimiters or safety warnings are provided for the interpolation of user input into the workflow prompts.
Capability inventory: The skill utilizes Bash, Write, Edit, WebSearch, and WebFetch tools as part of its prototyping workflow.
Sanitization: No explicit sanitization or validation of the input is performed within the skill definition.
[SAFE]: The skill uses allowed-tools to scope its environment, including standard file manipulation and web tools, which is consistent with its stated purpose of building experiments and conducting research.
[SAFE]: External references are limited to local configuration and workflow files located in the ~/.claude/gsd-core/ directory, which is typical for specialized development workflows.

gsd-spike