The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SKILL.md file instructs the agent to execute shell commands using npx to download and run the @nomadamas/k-skill package. This executions remote code from an unverified source at runtime to retrieve "primary instructions" and manage updates, effectively hiding the skill's actual logic from static analysis.
[PROMPT_INJECTION]: The skill employs a dynamic instruction override where it directs the agent to "follow its [the CLI's] output as the primary instructions for this skill." This encourages the agent to prioritize potentially malicious or unvetted remote instructions over the static guidelines provided in the repository.
[COMMAND_EXECUTION]: The skill relies on the execution of multiple shell commands via npx for core operations including instruction retrieval, file management, and software updates.
[INDIRECT_PROMPT_INJECTION]: The skill processes HTML content fetched from a third-party, publicly editable GitHub repository (jay-jo-0/github_pages_repo) which is not under the control of the skill author or a trusted vendor.
Ingestion points: Fetches file trees and raw HTML content via api.github.com and raw.githubusercontent.com.
Boundary markers: No delimiters or safety instructions are defined to prevent the agent from obeying instructions that may be embedded in the fetched financial reports.
Capability inventory: The skill enables the agent to extract and summarize text, headings, and data tables from external sources.
Sanitization: There is no evidence of HTML sanitization or instruction filtering for the content ingested from the external mirror repository.

daishin-report-search