The Agent Skills Directory

[PERSISTENCE]: The skill instructs the agent to create a persistence mechanism by adding a cron job to the user's crontab. The command */5 * * * * ~/.claude/scripts/cron-bot-tasks.sh >> ~/.claude/logs/cron-bot-tasks.log 2>&1 ensures the task-processing script runs automatically every five minutes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from GitHub issue and PR comments, which creates a significant injection surface.
Ingestion points: The request field (comment body) collected from GitHub notifications via the bot-tasks.sh script.
Boundary markers: The skill lacks explicit delimiters or instructions to treat the comment body as untrusted data, relying only on a mention filter.
Capability inventory: The skill spawns sub-agents with broad permissions, including cloning repositories, implementing code changes, running tests, committing code, and pushing to remote repositories.
Sanitization: There is no evidence of sanitization or safety filtering for the external comment content before it is used to define agent tasks.
[COMMAND_EXECUTION]: The skill executes local shell scripts located in ~/.claude/scripts/, specifically bot-tasks.sh and cron-bot-tasks.sh, and invokes a /setup-scripts tool to regenerate them if missing.
[DYNAMIC_EXECUTION]: The skill utilizes a "team" pattern, dynamically spawning up to five concurrent sub-agents to perform complex code modifications and GitHub operations based on the parsed tasks.

bot-tasks