The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the creation and persistence of new skill definitions (SKILL.md) within the vendor's application directory (~/.opensquilla/proposals/). It also includes a runtime_e2e step which performs a controlled execution of the candidate skill to verify functionality. These actions are governed by multiple safety gates, including static analysis (linting) and a manual acceptance process.
[DATA_EXFILTRATION]: The workflow incorporates a history exploration step using the history-explorer skill to identify common usage patterns. This data is used solely to inform the design of the new meta-skill and is processed within the internal LLM pipeline.
[PROMPT_INJECTION]: The skill uses xml_escape and character truncation to sanitize user-provided messages and system prompts before they are interpolated into the LLM instructions for intent clarification and assembly. This reduces the risk of user-supplied text overriding the skill's logic.
[DATA_EXPOSURE]: As part of its design, the skill provides a risk_classify step that explicitly identifies and flags potential security concerns in generated proposals, such as network access, file system operations, and shell command execution.

meta-skill-creator