The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes tools such as codex exec and mcp__codex__codex to run model-generated scripts (Python, Node.js, and Bash) against a local corpus. This is an intended functionality for the 'Inductive' verification phase and is explicitly limited to a read-only sandbox environment to prevent unauthorized file modifications.
[PROMPT_INJECTION]: By processing external data sources (logs, datasets, code), the skill is subject to Indirect Prompt Injection. This is mitigated by the skill's design principles, which instruct the agent to use programmatic tools like ripgrep and scripts for analysis instead of directly ingesting raw untrusted text into the primary reasoning context.

Dialectic Loop