The Agent Skills Directory

[SAFE]: The skill implements log analysis logic through local file and database parsing. It follows security best practices such as read-only database access and robust error handling during file reads. The skill contains no network logic or persistence mechanisms.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted log data which presents an indirect injection surface. However, the risk is addressed through the use of summary reporting and normalization of sensitive or dynamic fields (such as URLs, IP addresses, and IDs) which prevents raw log content from directly influencing agent behavior.
Ingestion points: Log files and SQLite databases (state.db) accessed via scripts/analyze_logs.py, scripts/hermes_log_window.py, and scripts/state_failures.py.
Boundary markers: The output is presented in structured JSON or Markdown reports.
Capability inventory: Capabilities are restricted to local file read/write operations; no network or arbitrary execution tools are present in the core scripts.
Sanitization: Log normalization replaces potentially malicious or variable strings with static placeholders.
[COMMAND_EXECUTION]: Subprocess calls identified in tests/test_analyze_logs.py and tests/test_state_failures.py are strictly for testing the skill's own command-line interface in a controlled environment and do not constitute a runtime security risk.

log-analyzer