The Agent Skills Directory

[Privilege Escalation] (HIGH): The skill extensively uses sudo for system management, giving the agent administrative control over the host.
Evidence in REFERENCE.md: The deployment workflow includes sudo apt install, sudo pip3 install, and sudo systemctl commands.
[Persistence Mechanisms] (HIGH): The skill creates automated tasks that persist across sessions.
Evidence in REFERENCE.md: Instructions include setting up a recurring backup task via echo "..." | sudo crontab -.
[Indirect Prompt Injection] (HIGH): The skill processes data from database outputs while possessing high-privilege system capabilities, without boundary markers or sanitization.
Ingestion points: Database query results, EXPLAIN ANALYZE plans, and log files in EXAMPLES.md and REFERENCE.md.
Boundary markers: Absent; no delimiters are used to separate untrusted database content from agent instructions.
Capability inventory: Root-level execution via sudo, file deletion via rm -rf, and network access via aws s3.
Sanitization: Absent; variable paths and database-derived strings are used directly in shell commands.
[Command Execution] (HIGH): Example scripts utilize destructive commands on dynamic paths, posing a risk of arbitrary file deletion.
Evidence in EXAMPLES.md: The backup verification script executes rm -rf $TEST_DIR. If $TEST_DIR is manipulated or incorrectly defined, it can lead to system-wide data loss.
[Unverifiable Dependencies & Remote Code Execution] (MEDIUM): The skill performs unpinned and unverified package installations at runtime.
Evidence in REFERENCE.md: The installation steps use sudo pip3 install patroni[etcd], which fetches code from public repositories without integrity checks.

database-administrator