The Agent Skills Directory

[COMMAND_EXECUTION] (CRITICAL): Shell Command Injection vulnerability in the 'Output and Save' step. The skill instructs the agent to execute echo "<generated_prompt>" | python scripts/save_prompt.py <project_root> <phase_number> <phase_name>. Since <generated_prompt>, <project_root>, and <phase_name> are constructed from user-provided input without sanitization, an attacker can inject shell metacharacters (e.g., ;, &, |, or backticks) to execute arbitrary commands.
Evidence (SKILL.md): echo "<generated_prompt>" | python scripts/save_prompt.py <project_root> <phase_number> <phase_name>
[REMOTE_CODE_EXECUTION] (CRITICAL): The command injection vulnerability mentioned above directly facilitates Remote Code Execution (RCE), as the agent executes shell-level instructions on the host machine.
[PROMPT_INJECTION] (HIGH): Category 8: Indirect Prompt Injection. The skill's core function is to generate instructions for downstream subagents.
Ingestion points: Processes user-provided variables like PHASE_NAME and PHASE_DOC_PATH and reads references/implementation-prompt-template.md.
Boundary markers: Absent. The template uses simple curly-brace interpolation ({PLACEHOLDER}), which does not prevent malicious instructions in the input from overriding the template's intent.
Capability inventory: Possesses file-writing capabilities and local script execution (scripts/save_prompt.py).
Sanitization: None. The skill blindly interpolates user input into a prompt that directs orchestration logic, allowing an attacker to hijack the behavior of subsequent agent sessions.

prompt-generator