The Agent Skills Directory

[PROMPT_INJECTION]: Documentation in references/01-original-scaffold.md identifies that the reference file references/original/claude_code_deep_planning.txt contains hidden or bidirectional Unicode text, which can be used to hide instructions from the user while influencing the AI agent. Additionally, the skill's primary function involves processing untrusted repository data, which creates an attack surface for indirect prompt injection.
[REMOTE_CODE_EXECUTION]: The scripts/test_doctor.py script performs dynamic code loading of the doctor.py utility via importlib. This pattern of executing code from computed paths at runtime is a risk factor for unauthorized execution.
[COMMAND_EXECUTION]: The skill's instructions in SKILL.md and associated references direct the agent to execute a local Python script, scripts/doctor.py, for package validation and capability checks.
[EXTERNAL_DOWNLOADS]: The skill architecture is designed around several external, third-party dependencies for core reasoning tasks (e.g., mattpocock/skills, 6missedcalls/ultraplan, and obra/superpowers). These represent unverifiable dependencies that are integrated into the agent's workflow.

agent-project-orchestrator