The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs bash commands using heredocs (<<'EOF') to pass requirements and feature names to the codeagent-wrapper utility. If a user provides an input containing the string EOF, they can terminate the heredoc prematurely and execute arbitrary shell commands on the host system.
[PROMPT_INJECTION]: The skill's instructions include 'CRITICAL CONSTRAINTS' that explicitly claim to 'override all other instructions' and hold 'HIGHEST PRIORITY'. This pattern is a form of internal prompt injection used to bypass standard operational guidelines in favor of the skill's specific logic.
[REMOTE_CODE_EXECUTION]: The skill relies on a non-standard external utility called codeagent-wrapper for all code modifications. This utility serves as a proxy for remote execution across various LLM backends, but its source and safety cannot be verified within the skill context, representing an unverified dependency.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its automated workflow.
Ingestion points: User requirements are collected via AskUserQuestion in SKILL.md (Step 1).
Boundary markers: None present; user inputs are directly interpolated into bash commands and markdown templates.
Capability inventory: The agent has access to the Bash tool for command execution, the Write tool for file modification, and the codeagent-wrapper for parallel code execution.
Sanitization: No input validation or escaping is performed on requirements before they are used by the dev-plan-generator agent to create 'Test Commands' in the dev-plan.md file, which are subsequently executed in Step 4.

dev