agent-development — agentic threat model

Not certain from the listing — relies on the underlying Claude Code model execution environment. The primary threat is prompt injection during the generation of system prompts or frontmatter, which could lead to downstream agent hijacking.

Operates on local agent definition files, reference documents, and schema templates. Risks include local file path traversal or reading malicious agent definitions designed to exploit the parser.

The skill orchestrates the validation of other agents. Vulnerabilities could exist in how the framework parses agent frontmatter, tool allowlists, and triggering conditions, potentially leading to logic bypasses.

Includes a 'validate-agent.sh' script executed via bash. This introduces a direct command execution vector if the script does not safely sanitize inputs or if the local environment lacks proper sandboxing.

Not certain from the listing — there is no explicit mention of runtime observability, logging, or guardrails for the validation script other than basic linting outputs.

Not certain from the listing — compliance and access controls depend entirely on the host system running Claude Code. No built-in authentication or policy enforcement is described.

Directly influences the agent ecosystem by defining subagent behaviors, triggering conditions, and tool allowlists. A compromise here could result in the creation of malicious subagents that abuse agent-to-agent trust.