The Agent Skills Directory

[METADATA_POISONING]: The skill's display output instructions contain a link to github.com/gannnonh/kata, which uses a misspelled version of the author's username (gannnonh instead of gannonh). This matches patterns associated with typosquatting and could lead users to a deceptive or malicious fork of the repository.
[INDIRECT_PROMPT_INJECTION]: The skill fetches a remote CHANGELOG.md file and instructs the agent to extract and format its contents. This creates a surface for indirect prompt injection if the remote file were to contain malicious instructions.
Ingestion points: Remote file fetch from GitHub using the WebFetch tool in SKILL.md.
Boundary markers: None specified in the instructions to the agent for processing the fetched content.
Capability inventory: Shell command execution (cat), file system access via the kata-lib.cjs Node.js script (reading/writing config.json), and network access via WebFetch.
Sanitization: No sanitization is performed on the fetched content before it is parsed and displayed to the user.
[EXTERNAL_DOWNLOADS]: Fetches a CHANGELOG.md file from the author's legitimate GitHub repository (gannonh/kata) to perform version comparison.
[COMMAND_EXECUTION]: Executes shell commands to read the local VERSION file and includes a Node.js utility script (scripts/kata-lib.cjs) that performs file system operations, directory traversal, and configuration management.

kata-whats-new