The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to execute code from a remote source via npx @airshiplabs/cli, which fetches and runs a package without verification of the package author or contents (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill depends on unverified third-party packages such as @airshiplabs/cli and opencode-ai from the NPM registry, and the Homebrew tap sst/tap/opencode (SKILL.md, references/commands.md).
[CREDENTIALS_UNSAFE]: The documentation references paths to sensitive authentication files such as ~/.claude and ~/.codex/auth.json, and environment variables like AWS_ACCESS_KEY_ID, which increases the risk of credential discovery and harvesting (references/commands.md).
[COMMAND_EXECUTION]: The CLI tool supported by this skill includes an --exec flag that allows the execution of arbitrary shell commands as part of the dev server lifecycle, providing a vector for command injection (references/commands.md).
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection where data from external HTML can influence code generation. Ingestion points: HTML elements selected via the Airship visual proxy (SKILL.md). Boundary markers: None identified. Capability inventory: The agent has Bash, Read, Write, and Edit tools enabled, and the CLI allows arbitrary execution (SKILL.md, references/commands.md). Sanitization: No sanitization of the ingested UI data is described.

airship