The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates a build.sh script to automate document compilation using Pandoc. It also recommends the installation of system utilities such as pandoc, tmux, and tectonic through standard package managers like brew, apt, or pkg for Termux.- [DYNAMIC_EXECUTION]: The skill dynamically generates a bash script (finishbook.sh) to facilitate unattended chapter generation. This script launches a new agent session using the --permission-mode bypassPermissions flag. The skill instructions require the agent to warn the user about the security implications and explicitly forbid the agent from executing the script automatically.- [PRIVILEGE_ESCALATION]: The skill utilizes chmod +x to make dynamically generated scripts (build.sh, finishbook.sh) executable. The inclusion of the bypassPermissions flag in generated templates is a form of capability escalation that allows the agent to perform actions without individual confirmation prompts, although this is gated by manual user execution.- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided inputs to personalize book content and calibrate learning assessments, creating a surface for indirect injection.
Ingestion points: User data is collected and stored in learning/profile.md and learning/assessment.md.
Boundary markers: The skill uses structured Markdown templates but lacks explicit security delimiters to prevent user-supplied strings from influencing agent behavior during calibration steps.
Capability inventory: The agent has the ability to write files to the project directory and execute generated shell scripts.
Sanitization: The generated script includes basic sanitization for tmux session names using the tr command to ensure the identifier is safe for shell use.

helpmelearn