The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted input such as prose, code trees, or repository layouts to generate diagrams. This creates a surface for indirect prompt injection where malicious instructions embedded in the data could influence the agent's rendering logic. \n
Ingestion points: SKILL.md identifies input as prose, package trees, or repo layouts. \n
Boundary markers: No specific delimiters or "ignore" instructions are defined for the input data to prevent the agent from following instructions found within the data. \n
Capability inventory: The skill utilizes scripts/tq to execute JavaScript code on a local tldraw application and scripts/new-canvas to manage local files and applications. \n
Sanitization: No evidence of sanitization or validation of the input data before processing.\n- [COMMAND_EXECUTION]: The skill uses local shell scripts (scripts/new-canvas, scripts/tq) to perform its tasks. These scripts invoke system commands like curl, python3, mkdir, cp, and open to interact with the local environment and the tldraw application.\n- [SAFE]: The skill implements idempotency by tagging its created shapes with meta.kckit metadata. This ensures that repeated runs of the skill update existing diagrams rather than creating duplicate shapes, protecting user-edited content.\n- [SAFE]: All network operations are directed at localhost to communicate with the local tldraw server bridge. There is no evidence of communication with external, untrusted domains.

package-diagram