The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to clone a repository from an untrusted external source (https://github.com/jnMetaCode/agency-agents-zh.git) and execute local shell scripts (./scripts/install.sh, ./scripts/convert.sh) which can perform arbitrary operations on the host system.
[EXTERNAL_DOWNLOADS]: The skill requires the global installation of a third-party package (agency-orchestrator) via npm install -g, introducing external code dependencies that are not from a verified or trusted organization.
[COMMAND_EXECUTION]: The instructions encourage the execution of several shell commands, including git, npm, and custom binaries (ao), which interact with the file system and external networks to perform multi-agent orchestration tasks.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface for indirect prompt injection.
Ingestion points: Natural language input to the ao compose command and the task fields within workflow.yaml files.
Boundary markers: The provided examples do not demonstrate the use of delimiters or instructions to ignore embedded malicious content within tasks.
Capability inventory: The agents have the capability to generate code, design architectures, and execute complex multi-step workflows across various departments (engineering, marketing, security).
Sanitization: There is no evidence of sanitization or validation of the input data before it is processed by the agent personas.

agency-agents-zh-skill