The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local TypeScript tool (Tools/Tldr.ts) using the Bun runtime to perform file operations on .tldr canvases. These operations include creating, inspecting, and mutating canvas records.
[DATA_EXFILTRATION]: The skill uses curl to send status notifications to a local endpoint (http://localhost:31337/notify). While this is a network operation targeting localhost, it represents a side-channel communication mechanism.
[PROMPT_INJECTION]: The StructureCanvas workflow processes external, user-provided .tldr files which contain hand-drawn text and notes. This constitutes an indirect prompt injection surface where malicious instructions within a canvas could attempt to influence agent behavior.
Ingestion points: The Tldr.ts inspect command reads record data (text, labels, titles) from local .tldr files in the StructureCanvas.md workflow.
Boundary markers: The instructions do not define specific delimiters or "ignore" warnings when presenting the inspected JSON data to the model for organization.
Capability inventory: The skill possesses capabilities to write to the local file system (Tldr.ts add/move/settext) and execute shell commands via Bun.
Sanitization: The Tldr.ts tool performs schema and structure validation but does not sanitize or filter the content of the richText properties within the canvas records.
[PROMPT_INJECTION]: The skill instructions direct the agent to load and apply configuration from ~/.claude/LIFEOS/USER/CUSTOMIZATIONS/SKILLS/Tldraw/PREFERENCES.md if it exists. This file is an external ingestion point that could be used to inject instructions into the skill's setup phase.

Tldraw