The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes pnpm install (Step 2) in a directory determined by user input. This action triggers Node.js lifecycle scripts (preinstall, postinstall) which can execute arbitrary code contained within the target package's manifest. Further execution of dsh plugin add with GitHub URLs (Step 6) may also involve downloading and running remote code via build scripts.- [COMMAND_EXECUTION]: The workflow relies heavily on executing shell commands such as pnpm, git, npm, and dsh. Many arguments for these commands, including package names, versions, and repository paths, are derived from external files in the project root without explicit validation.- [DYNAMIC_EXECUTION]: The skill executes scripts defined in a local package.json file, specifically pnpm run build and pnpm run lint (Step 2). The behavior of these commands is entirely dependent on the configuration within the untrusted local source.- [DATA_EXFILTRATION]: The skill is designed to transmit local code to external services via pnpm publish and git push. While intentional, the skill is instructed to proceed without asking the user for channel selection if none is provided, which could lead to unintentional publication of sensitive data to public registries like npm.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data that could influence its behavior.\n
Ingestion points: Reads package.json and dsh.bundle.patch from the resolved plugin source directory.\n
Boundary markers: None identified; instructions do not include delimiters or warnings to ignore embedded directives in the ingested manifests.\n
Capability inventory: Includes high-privilege operations like package installation, publishing to npm, and pushing to git repositories.\n
Sanitization: There is no evidence of sanitization or schema validation for values extracted from project manifests before they are interpolated into shell commands.

dsh-plugin-publisher