The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is installed using npx from the author's GitHub repository.
[COMMAND_EXECUTION]: The scripts/dev-agent.py script manages the autonomous development loop by executing shell commands via the subprocess module. In 'Autopilot' mode, it spawns claude processes with the --dangerously-skip-permissions flag, which bypasses security prompts for tool execution.
[REMOTE_CODE_EXECUTION]: The 'Initializer' agent generates a shell script (init.sh) based on the project specification provided by the user. The workflow then requires the agent to execute this dynamically created script. This allows for arbitrary code execution if the project specification contains malicious instructions.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because it processes untrusted input from project specifications and source files to generate executable tasks and scripts.
Ingestion points: Project specification files (e.g., app_spec.txt, spec.md) and the project's source code.
Boundary markers: Absent; no delimiters are used to separate user-provided content from instructions.
Capability inventory: subprocess.run and subprocess.Popen in scripts/dev-agent.py; file writing for project initialization.
Sanitization: Absent; the generated init.sh and feature_list.json are not sanitized or validated.
[PROMPT_INJECTION]: The system prompts in SKILL.md and AGENTS.md use rigid, imperative language that could be exploited to bypass safety guardrails if the agent's context is poisoned by malicious project data.

infinite-dev