The Agent Skills Directory

[SAFE]: The skill instructions and reference materials do not contain any malicious patterns. It follows security best practices by enforcing a clear separation between graph composition (non-charging) and graph execution (charging), and explicitly forbids the use of flags that would allow the agent to bypass user approval for credit spending.
[COMMAND_EXECUTION]: The skill documents the use of the dreamina-canvas CLI tool for creating and editing canvas nodes. These operations are restricted to non-executing commands (create, edit, find, show) and do not involve arbitrary or dangerous command execution beyond the defined scope of the tool.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the outputs of the dreamina-canvas tool (specifically from node find and node show).
Ingestion points: Data returned from CLI summary and detail views in SKILL.md.
Boundary markers: The skill uses explicit instructions to treat returned IDs as authoritative and validates bindings against a live schema.
Capability inventory: Limited to node structure management via the dreamina-canvas CLI; no shell execution, file-writing, or network operations are present in this skill.
Sanitization: Relies on internal CLI validation and schema checks mentioned in the instructions.

dreamina-canvas-compose