The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by design as it ingests untrusted data from an external source.
Ingestion points: Untrusted content from cubic.ai is accepted via the $ARGUMENTS variable in SKILL.md.
Boundary markers: While the skill uses <cubic-analysis> tags within its own instructions, the content is subsequently interpolated into a natural language task description string ([ANALYSIS]) for the gsd:quick tool without escaping or strict delimiters to prevent the agent from following instructions embedded within that content.
Capability inventory: The skill invokes the gsd:quick skill (SKILL.md), which belongs to an autonomous workflow capable of reading the codebase and applying modifications.
Sanitization: No sanitization, validation, or filtering is applied to the cubic.ai output before it is passed to the powerful gsd:quick tool.
[EXTERNAL_DOWNLOADS]: The skill documentation and configuration involve external resources.
The README.md provides an installation command using npx that fetches from the author's own repository (kwazema/claude-skills).
The skill requires the Get Shit Done (GSD) workflow, which is an external dependency hosted at github.com/get-shit-done-ai/get-shit-done.

kw-cubic