The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes logs from various sources which can contain untrusted data.
Ingestion points: Discovers and reads log files in logs/, var/log/, and via service managers like Docker (docker compose logs), Systemd (journalctl), and PM2.
Boundary markers: The skill does not define clear boundaries or provide instructions to ignore embedded commands within the log data during the analysis phase.
Capability inventory: The skill utilizes the Edit tool for file modification and Bash for command execution (e.g., running tests, docker commands) during the fix and validation phases.
Sanitization: No explicit sanitization or filtering of log content is described before it is used to inform root cause analysis and code fixes.
[COMMAND_EXECUTION]: The skill executes various shell commands based on the detected environment and log analysis results.
Evidence: Uses docker compose, journalctl, and pm2 for log source discovery.
Evidence: Executes validation commands like pytest and npm test after applying fixes.
[DYNAMIC_EXECUTION]: The skill uses dynamic interpreter calls to determine file formats.
Evidence: Executes a dynamic Python snippet via python3 -c to check if a log line is valid JSON.

sc-log-fix