The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill preamble contains instructions to execute the output of the local gstack-slug command using shell eval. This pattern allows the framework to dynamically inject environment variables or shell commands into the agent's active execution context based on tool output.
[DYNAMIC_EXECUTION]: The skill is designed to process and act upon dynamic GSTACK_INSTRUCTION blocks returned by the gstack-skill-start tool. This mechanism facilitates the runtime delivery of instructions to the agent from an external binary.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting instructions from tool outputs. The instruction set includes specific boundary markers (GSTACK_INSTRUCTION_BEGIN and GSTACK_INSTRUCTION_END) and requires the agent to validate the source tool and SESSION_ID before execution to mitigate cross-source injection risks.
Ingestion points: Standard output of the gstack-skill-start binary.
Boundary markers: Defined start and end delimiters for instruction blocks.
Capability inventory: Access to Bash, Read, Edit, Glob, Grep, and AskUserQuestion tools.
Sanitization: Verification of SESSION_ID and source tool identity before processing dynamic instructions.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run several local scripts and binaries within the ~/.claude/skills/gstack/bin/ directory. These tools manage internal framework state, telemetry, learning logs, and project metadata stored in ~/.gstack/.

ios-clean