The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides installation instructions for the Crossplane CLI that download a script from https://raw.githubusercontent.com/crossplane/crossplane/master/install.sh and pipe it directly to the system shell (sh). This practice allows for arbitrary code execution from a remote source and is a major supply chain risk.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to fetch external software components using curl and helm during the setup process.
[COMMAND_EXECUTION]: The skill uses the sudo command to move a binary to a restricted system path (/usr/local/bin/), which represents a privilege escalation finding.
[PROMPT_INJECTION]: The skill exhibits surface area for indirect prompt injection. 1. Ingestion points: The skill uses kubectl logs and kubectl describe in SKILL.md to read potentially untrusted data from the Kubernetes cluster. 2. Boundary markers: No delimiters or isolation instructions are used to separate this external data from the agent's instructions. 3. Capability inventory: The skill is granted access to the Bash tool, providing powerful system-level interaction. 4. Sanitization: No evidence of sanitization or validation of the ingested data is present in the skill instructions.

loom-crossplane