The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions explicitly direct the agent to use an "unsandboxed shell" to execute loadout sync --global because the command "needs to write outside ~/wrksp". This is a direct instruction to the AI to bypass platform-enforced security boundaries and workspace restrictions.
[PERSISTENCE]: The skill modifies shell initialization files (.airc.d/claude.zsh, .airc.d/opencode.zsh) and instructs the agent to source ~/.airc. Modifying shell profiles allows for the insertion of arbitrary commands that execute every time a terminal session is initiated.
[COMMAND_EXECUTION]: The skill executes multiple system-level commands and external tools, including loadout, pi, prat, and shell built-ins like source and grep on configuration files.
[DYNAMIC_EXECUTION]: The instructions in SKILL.md contain a Python snippet that uses importlib.util and exec_module to dynamically load and execute the skill's own rank_models.py script at runtime.
[EXTERNAL_DOWNLOADS]: The rank_models.py script fetches data from Artificial Analysis and OpenRouter to determine model IDs and intelligence metrics used in configuration updates.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted data from external websites and uses that data to overwrite system configuration and shell profile files.
- Ingestion points: The script rank_models.py fetches and parses HTML/JSON payloads from artificialanalysis.ai/leaderboards/models and artificialanalysis.ai/models.
- Boundary markers: Absent. The skill does not use delimiters or provide warnings to the agent regarding instructions that might be embedded in the fetched model names or metadata.
- Capability inventory: The skill has the capability to write to shell profiles (~/.airc.d/), update global loadout settings, and execute shell commands via an unsandboxed environment.
- Sanitization: Data processing is limited to regex parsing and JSON extraction; there is no evidence of sanitization or escaping of the external data before it is interpolated into shell scripts.

agent-models