The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to use npx -y lavish-axi to download and execute a package from the npm registry if not already installed locally.
[REMOTE_CODE_EXECUTION]: Executing code via npx from the npm registry constitutes remote code execution from a third-party source.
[COMMAND_EXECUTION]: The skill invokes several shell commands to manage review surfaces, including lavish-axi poll, lavish-axi export, and lavish-axi share.
[DATA_EXFILTRATION]: The lavish-axi share command facilitates uploading local artifacts to the external domain ht-ml.app. The skill includes a specific mitigation requiring explicit user approval before every upload.
[DYNAMIC_EXECUTION]: The skill executes Node.js scripts from paths computed at runtime, such as node "$(npm root)/lavish-axi/dist/cli.mjs".
[INDIRECT_PROMPT_INJECTION]: The skill polls for external feedback data which is then applied by the agent, creating a surface for untrusted instructions.
Ingestion points: Output from the lavish-axi poll command and associated Excalidraw/PNG feedback files.
Boundary markers: A one-shot review mode is the default setting, which limits the interaction to a single batch of feedback.
Capability inventory: Shell execution of the lavish-axi CLI, local file system writes to the .lavish/ directory, and network uploads via the share command.
Sanitization: No explicit sanitization or validation logic is defined for the content retrieved from external feedback polls.
[METADATA_POISONING]: The author listed in the skill metadata ('Kun Chen') does not match the expected author ('hexsleeves').

lavish