The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands via the jf CLI and npx @jfrog/agent-guard to manage lifecycle operations for skills and plugins.
[EXTERNAL_DOWNLOADS]: Downloads the @jfrog/agent-guard package from the JFrog releases registry (https://releases.jfrog.io/artifactory/api/npm/coding-agents-npm/) at runtime using npx.
[DYNAMIC_EXECUTION]: Executes external software and local scripts including npx @jfrog/agent-guard, a local check-environment.sh script, and harness-specific binaries (e.g., claude, cursor) for plugin management.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: Installs remote skill and plugin archives from Artifactory repositories. These operations are managed by the vendor CLI and include security controls like evidence verification and Xray scanning status checks.
[INDIRECT_PROMPT_INJECTION]: The skill processes metadata (names, descriptions, versions) for skills and plugins retrieved from remote JFrog repositories, which could contain malicious instructions.
Ingestion points: Data enters the context through npx @jfrog/agent-guard --list-skills and jf skills list outputs.
Boundary markers: The skill uses rigid Markdown table templates and explicit user confirmation prompts for all installations or removals.
Capability inventory: Capabilities include file system writes (jf install), directory deletion (rm -rf), and command execution (npx, bash).
Sanitization: Employs jq and awk for structured parsing of tool output to avoid direct interpolation of untrusted strings into the prompt context.
[DATA_EXPOSURE_AND_EXFILTRATION]: Accesses local configuration files such as ~/.jfrog/setup.json to resolve project identifiers. The skill explicitly instructs the agent not to read jfrog-cli.conf.v6 to prevent the exposure of sensitive access tokens.

jfrog-ai-catalog