The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates and system prompts that interpolate user-supplied intent and descriptions into agent configurations. While standard for this use case, it creates a potential surface for indirect prompt injection if the user-provided data contains instructions aimed at overriding the agent architect's persona.
Ingestion points: examples/agent-creation-prompt.md (user description interpolated into prompt), scripts/validate-agent.sh (processes user-provided agent files).
Boundary markers: The templates use structured JSON output formats and clear markdown headers to separate instructions from data.
Capability inventory: The validation script uses standard shell utilities (sed, grep, awk) for reading. Example agents described in the skill utilize Read, Write, Grep, and Bash tools.
Sanitization: No explicit sanitization or escaping of the user-provided description is described in the prompt templates.
[COMMAND_EXECUTION]: The skill includes a utility script scripts/validate-agent.sh for validating agent files.
Evidence: The script uses standard Unix utilities (head, tail, grep, sed, awk) to parse and validate the structure of markdown files provided as command-line arguments. The script uses proper quoting for variables (e.g., "$AGENT_FILE") to prevent simple shell injection.

agent-identifier