The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data and runtime logs without instructing the agent to use boundary markers or perform sanitization, which could allow malicious instructions embedded in those sources to influence agent behavior.
Ingestion points: Runtime log fetching (pre/runtime-logs.md, post/runtime-logs.md), web page retrieval and asset import (pre/web-assets.md, post/web-assets.md), and OpenAPI/Swagger documentation imports (pre/third-party-api.md).
Boundary markers: Absent. The agent is not guided to treat external content as data using delimiters.
Capability inventory: Shell command execution via npx, database mutations (runtime insert/update), and application deployment functions.
Sanitization: Absent.
[REMOTE_CODE_EXECUTION]: The skill instructions rely on downloading and executing the platform's CLI tool using npx at runtime.
Evidence: Frequent use of npx -y zion-mcp@2.7.13 throughout the skill (e.g., SKILL.md, pre/actionflow.md) for core operations.
Note: This is a vendor-owned resource (zion-mcp) from the author (functorz-tech) used for platform management.
[DYNAMIC_EXECUTION]: The skill documents the platform's ability to execute custom JavaScript in a server-side sandbox and via CLI commands.
Evidence: Detailed guidance on the "Run Code" node in pre/actionflow.md and post/actionflow.md, and the runtime run-code verb in pre/runtime-verify.md and post/runtime-verify.md.
[COMMAND_EXECUTION]: The skill is primarily driven through shell commands provided to the agent for managing the Zion platform infrastructure.
Evidence: Extensive use of CLI examples for project configuration, schema management, and runtime interaction across all capability files.

zion-platform