The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a template for and offers to generate an overnight.sh script that executes the agent with the --permission-mode bypassPermissions flag. This configuration explicitly disables the security guardrails that require human approval for shell commands, allowing for the autonomous execution of potentially harmful operations.
[DYNAMIC_EXECUTION]: The skill dynamically generates a shell script (overnight.sh) based on a template and the specific project context. It uses chmod +x to grant execution permissions to the generated file, facilitating its use as a standalone executable.
[INDIRECT_PROMPT_INJECTION]: The skill creates a workflow where an autonomous agent (using the bypassed permissions mentioned above) reads and follows instructions from the spec/ folder. This presents a high-risk surface for indirect prompt injection.
Ingestion points: spec/README.md, spec/structure.md, spec/roadmap.md, and project-specific files in spec/plans/.
Boundary markers: None identified; the agent is instructed to follow the spec files directly.
Capability inventory: The autonomous agent has full shell and filesystem access due to the bypassPermissions flag.
Sanitization: None; the agent processes the content of the spec files as authoritative build instructions.
[PRIVILEGE_ESCALATION]: The skill includes instructions to modify file permissions (chmod +x) and recommends using system package managers (brew, apt, pkg) to install utilities like tmux, which typically requires administrative or root privileges.

helpmeplan