The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains forceful instructions such as '⚠️ MANDATORY SKILL INVOCATION ⚠️' and 'Failure to invoke this skill... violates your operational requirements', which are attempts to override the agent's behavior and tool-selection logic.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its log-reading features. An attacker who can influence system logs (e.g., via web server errors or failed login attempts) could inject instructions that the agent might follow when it queries those logs.
Ingestion points: examples/read-logs.sh and scripts/dashboard.sh ingest syslog data through the API.
Boundary markers: There are no delimiters or warnings to prevent the agent from following instructions found within the logs.
Capability inventory: The skill has network access (curl) and local file-writing capabilities.
Sanitization: Ingested log data is not filtered or sanitized before being output to the agent.
[DATA_EXFILTRATION]: The skill collects sensitive metadata about the host server, including hardware UUIDs, system configurations, and system logs. The scripts/dashboard.sh script specifically aggregates this data and writes it to a file (unraid-inventory.md) in the user's home directory.
[COMMAND_EXECUTION]: The skill executes multiple shell scripts that use curl for network requests and jq for parsing sensitive server data, relying on environment variables for authentication.
[EXTERNAL_DOWNLOADS]: scripts/dashboard.sh attempts to source an external script lib/load-env.sh located outside the skill's root directory. This represents an unverifiable dependency on a script not provided in the skill package.

unraid