The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to construct shell commands using the $ARGUMENTS variable without explicit sanitization. Specifically, the instruction GRAPH_DIR=<project-dir> npx vite --host 127.0.0.1 uses the user-provided project path directly in an environment variable assignment, which could lead to command injection if a path containing shell metacharacters (e.g., ;, &&, `) is provided.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where untrusted data (project paths provided by the user or potentially found in a compromised repository) is interpolated into shell commands.
Ingestion points: $ARGUMENTS (user-supplied path) and filesystem checks for .understand-anything/knowledge-graph.json.
Boundary markers: None present in the instructions to warn the agent about malicious path strings.
Capability inventory: Uses bash tool to resolve paths, run pnpm install, pnpm build, and npx vite.
Sanitization: No sanitization or escaping logic is defined for the <project-dir> or <dashboard-dir> placeholders.
[DYNAMIC_EXECUTION]: The skill performs runtime installation and compilation of local packages using pnpm install --frozen-lockfile and pnpm --filter @understand-anything/core build. While these are standard development workflows, they involve executing code and build scripts residing in local directories that the agent is instructed to find via a broad search of the home directory.

understand-dashboard