The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of the arch-viz tool by cloning a repository from https://github.com/kplliaoshen666-droid/arch-viz-studio.git and running npm install. This source is not associated with the stated author or any known trusted organization.
[REMOTE_CODE_EXECUTION]: By instructing the agent to clone and install a third-party repository, the skill facilitates the execution of arbitrary installation scripts (e.g., npm postinstall). Furthermore, the --use-target-codegraph flag explicitly allows the tool to execute a local version of codegraph found within the repository being scanned, which represents a significant risk when processing untrusted codebases.
[COMMAND_EXECUTION]: The skill utilizes shell commands to manage tool installation (git clone, npm install, npm run install:global) and to perform the architecture scans (arch-viz scan).
[INDIRECT_PROMPT_INJECTION]: The tool ingests untrusted data by scanning repository source code to generate an architecture graph. This graph is then read by the agent to answer questions, creating a surface where malicious repository content could potentially influence agent responses.
Ingestion points: Local repository files processed during arch-viz scan, specifically results written to graph.json.
Boundary markers: None implemented for the resulting graph.json data when read by the agent.
Capability inventory: Subprocess execution for scanning, file write access to the docs/architecture/ directory.
Sanitization: No evidence of input validation or sanitization for the scanned repository content or the generated graph data.

arch-viz