The Agent Skills Directory

Indirect Prompt Injection Surface: The skill's primary function involves analyzing the current conversation transcript to extract requirements and end states. This creates a potential surface where external or user-provided data could influence the generated logic.
Ingestion points: The transcript of the live conversation is the primary input for Phase 1 of the skill.
Boundary markers: The skill does not implement specific delimiters or instructions to ignore directives found within the conversation history.
Capability inventory: The skill's workflow includes writing files to the .amplifier/goals/ directory and recommending the execution of the /goal command via the terminal.
Sanitization: The skill includes a 'Lint' phase that checks the generated condition against specific rules (e.g., ordering constraints, open enumerations). While focused on functional correctness, this serves as a validation mechanism for the output.
Command Generation and File Operations: The skill is designed to write goal conditions to local files and suggest the corresponding CLI command to the user. This approach ensures security by keeping the user in control, as the skill is explicitly directed to offer the command rather than auto-running it, facilitating manual review of the drafted file.

goalify