The Agent Skills Directory

[DYNAMIC_EXECUTION]: The json_get function in scripts/publish.sh uses node -e to evaluate code strings via the new Function constructor. This is used as a lightweight alternative to jq for parsing Cloudflare CLI output.
Evidence: const v=new Function("j","return "+process.argv[1])(JSON.parse(s)) in scripts/publish.sh.
[PRIVILEGE_ESCALATION]: The skill includes instructions to programmatically change Cloudflare R2 bucket permissions to enable public access using the --force flag, which lowers the security posture of the cloud storage.
Evidence: cf r2 buckets domains managed update plans --enabled true --force in SKILL.md and scripts/publish.sh.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to publish content that could be influenced by untrusted external data during the plan creation phase, potentially leading to the public exposure of unintended information.
Ingestion points: The HTML file content passed as the first argument to scripts/publish.sh.
Boundary markers: None. The instructions do not specify delimiters or warnings for the agent to ignore instructions embedded within the data it is turning into HTML.
Capability inventory: File upload via cf r2 objects put, network verification via curl, and dynamic JSON processing via node -e.
Sanitization: No sanitization or validation of the HTML content is performed prior to the public upload.
[COMMAND_EXECUTION]: The skill relies on the execution of shell scripts and the external cf CLI tool to perform its primary functions.
Evidence: Execution of scripts/publish.sh which wraps several cf r2 commands.

publish-plan