The Agent Skills Directory

[METADATA_POISONING]: The skill's YAML frontmatter lists the author as 'clerk', which contradicts the established authorship attribution of 'delorenj'. This deceptive metadata could mislead users or agents regarding the skill's official status and provenance.
[INDIRECT_PROMPT_INJECTION]: The workflow requires the agent to find a documentation URL in a local README file, append '.md', and then 'visit and read that markdown URL to compile a required-step checklist'. This pattern ingests remote, untrusted data directly into the agent's task-planning context without boundary markers or sanitization, allowing content in the remote file to potentially override agent instructions.
Ingestion points: Remote markdown URL fetched via WebFetch based on local package contents (SKILL.md, references/custom.md, references/prebuilt.md).
Boundary markers: Absent; the agent is instructed to treat the fetched content as a source of truth for implementation steps.
Capability inventory: Modifies project files (.xcodeproj), installs packages, enables app capabilities, and configures Associated Domains.
Sanitization: None provided for the remote markdown content.
[EXTERNAL_DOWNLOADS]: The skill utilizes the WebFetch tool to access the '/v1/environment' endpoint and remote documentation pages on clerk.com and GitHub. While these are targeted at legitimate Clerk service domains, they involve the processing of external data to drive local project changes.
[COMMAND_EXECUTION]: The skill automates project modifications, including the installation of the 'clerk-ios' Swift package and the addition of app capabilities and Associated Domains (e.g., 'webcredentials:{YOUR_FRONTEND_API_URL}') to the Xcode project configuration.

clerk-swift