The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to perform git operations and create pull requests via the GitHub CLI (gh). It interpolates data from external design bundles (such as the PR title) directly into these shell commands. This creates a potential surface for command injection or execution errors if the input bundle contains malicious or malformed shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill operates as an orchestrator that ingests external design data and passes it to sub-agents (ork:test-generator, ork:expect-agent).
Ingestion points: The skill accepts a Claude Design handoff URL or a local JSON bundle file as input in SKILL.md.
Boundary markers: The instructions do not explicitly include delimiters or instructions to the sub-agents to ignore potentially malicious content embedded in the component code or manifest data.
Capability inventory: The skill possesses significant capabilities including Bash for shell access, Write for file modification, and the ability to spawn sub-agents with full tool access.
Sanitization: There is no evidence of sanitization or validation of the fields extracted from the bundle before they are used in prompts or shell commands.
[EXTERNAL_DOWNLOADS]: The skill utilizes the WebFetch tool to retrieve design bundles from claude.ai. This is an official domain for a well-known service and is a standard operation for the skill's primary purpose.

design-ship