The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frontmatter requests access to the Bash tool. This capability allows the underlying logic in the referenced workflow file to execute shell commands on the host system.
[DYNAMIC_EXECUTION]: The skill utilizes the execution_context field to point to a local file path at ~/.claude/gsd-core/workflows/resume-project.md. This indicates that the core operational logic is loaded dynamically from the file system at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and act upon instructions or data found within STATE.md and .continue-here files to resume work. This creates a surface where malicious instructions embedded in these project files could influence agent behavior.
Ingestion points: Reads project context from STATE.md and checkpoint files (.continue-here).
Boundary markers: None defined in the skill instructions to distinguish between trusted system state and potentially untrusted project content.
Capability inventory: Access to Bash, Read, Write, and Grep tools provides significant system interaction capabilities.
Sanitization: No explicit validation or sanitization of the contents of the state files is mentioned.

gsd-resume-work