The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes bash scripts (scripts/pb_healthcheck.sh and scripts/pb_request.sh) that execute shell commands using variables provided by the environment or agent instructions.
Evidence: curl -sS -X "$method" "$url" in scripts/pb_request.sh and curl -sS -o /tmp/pb-health.json in scripts/pb_healthcheck.sh.
Risk: Standard utility scripts that rely on the agent to correctly populate method and path arguments. Malicious manipulation of these arguments could lead to unintended network operations.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and displays raw data from external API responses.
Ingestion points: scripts/pb_healthcheck.sh and scripts/pb_request.sh print the raw JSON output from the PocketBase API to the console.
Boundary markers: Absent. No delimiters or instructions are provided to the agent to ignore instructions that might be embedded in the API response bodies.
Capability inventory: The skill has the capability to perform authenticated administrative requests and write to the local filesystem (/tmp).
Sanitization: Absent. The scripts do not filter or validate the content of the API responses before presenting them to the agent context.

pocketbase-operations