The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes dynamic context injection markers (!) in SKILL.md to execute environment checks for the claude CLI, jq, and the git repository root at load time. These are diagnostic checks to ensure prerequisites are met.
[COMMAND_EXECUTION]: The lane-launcher.sh script executes shell commands to manage background sessions, including starting sessions via claude --bg and stopping them via claude stop. These commands are standard for the skill's session management purpose.
[EXTERNAL_DOWNLOADS]: The skill performs repository updates via git pull and updates the plugin marketplace via claude plugin marketplace update. These actions target the local repository and the official plugin catalog.
[DATA_EXFILTRATION]: The telemetry-upsert.sh script uploads environment data to GitHub issues via the gh CLI. The script includes a critical security check to ensure that the file being uploaded resides within a designated 'safe' directory (e.g., $CLAUDE_PLUGIN_DATA), preventing the exfiltration of arbitrary sensitive files from the host system.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it reads content from local files (defined in lanes.json in lane-launcher.sh) to seed background agent sessions. These sessions have background processing capabilities, and the prompt files are read without sanitization or boundary markers. However, this is the intended primary purpose of the skill and relies on trusted local configuration.

lanes