The Agent Skills Directory

[COMMAND_EXECUTION]: The provided Bash watcher script uses curl to measure response times of user-supplied targets. The $target variable is used without quoting or sanitization, which creates a command injection surface if the script is invoked with malicious inputs (e.g., a target containing shell metacharacters like semicolons or backticks).
[DYNAMIC_EXECUTION]: The skill uses a Bash heredoc pattern (`python3
<< 'PY'`) to execute inline Python code at runtime. While this is used for processing local log files and calculating deltas, it represents a dynamic execution pattern where code is generated and run in a single step.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external sources, such as HTTP response headers, body sizes, and authentication status codes (e.g., M365 lockout counts). This creates a surface where a monitored target could potentially influence the agent's logic or findings by manipulating these signals.
Ingestion points: HTTP response data (timing, size, headers, cookies) via curl and lockout status from o365_results.jsonl.
Boundary markers: None identified in the provided templates or scripts to distinguish untrusted target data from agent instructions.
Capability inventory: The skill performs network requests (curl) and persistent local file writes (echo >> and JSON logging).
Sanitization: The provided scripts do not perform validation or sanitization on the external data before logging it or processing it through the Python diffing logic.

mid-engagement-ir-detection