The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources such as workspace manifests (e.g., package.json, requirements.txt) and local configuration files. To mitigate injection risks, the instructions mandate using a dedicated validation script, scripts/check-repo-key.sh, which enforces a strict allowlist of characters for all repository keys and server IDs before they are used in commands.
Ingestion points: Project manifest files and JSON configuration files located at .jfrog/local/package-resolution.json and ~/.jfrog/skills-cache/package-resolution.json.
Boundary markers: The skill uses shell variables and quoted heredocs to ensure that external data is treated as literal strings and not executable code.
Capability inventory: The skill executes shell commands using the jf CLI and local utility scripts for configuration management.
Sanitization: All user-controlled or file-sourced inputs used in shell commands are passed through a validation script that rejects any input not matching the regex ^[A-Za-z0-9._-]+$.
[DYNAMIC_EXECUTION]: The skill includes a mechanism to refresh the routing policy by locating and executing a specific script (print-policy.mjs) found in the agent's plugin directory.
Evidence: Step 4 of the main instructions uses a find command to locate the policy script within established agent plugin folders and executes it using node.
Safety measures: The process includes path validation checks to ensure the script is located outside the workspace and that the path contains only safe characters.

jfrog-setup-package-managers