The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from log files, raw text, or external URLs, which provides a surface for indirect prompt injection attacks.
Ingestion points: The INPUT_SOURCE variable in SKILL.md (Step 1) accepts file paths, URLs, or raw text input.
Boundary markers: Absent. The instructions do not include delimiters or specific guidance for the agent to ignore instructions that might be embedded within the analyzed logs or website content.
Capability inventory: The skill utilizes shell commands (cat, curl, pip), executes Python scripts via heredocs, and performs file writes to the /tmp directory (Steps 1-5).
Sanitization: Absent. The content is processed directly by regex and library-based extractors without pre-processing to remove potential injection payloads.
[DATA_EXFILTRATION]: The skill provides a mechanism to read local files (e.g., system logs or configuration files) via the INPUT_SOURCE parameter. While intended for security analysis, this capability could be used to access sensitive files if an attacker provides a path like /etc/passwd. Additionally, the use of curl in Step 1 allows the skill to send requests to arbitrary user-supplied domains.
[COMMAND_EXECUTION]: The workflow relies on shell command execution (cat, curl, wc) and uses Python heredocs (`python3
<<'PYEOF'`) to execute logic, which is a common but powerful capability that should be monitored.
[EXTERNAL_DOWNLOADS]: The skill uses curl -sL to fetch content from remote URLs provided by the user and suggests the installation of the ioc-finder package from the public PyPI registry.

ioc-extract