The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and applies diff-based patches to SKILL.md files. Since these files define the agent's operational logic and constraints, this constitutes a dynamic self-modification of the agent's instruction set based on runtime analysis.
[DATA_EXFILTRATION]: The skill reads .aris/meta/events.jsonl, which contains sensitive interaction metadata including previews of user prompts and tool usage summaries. This data is transmitted to an external model (gpt-5.4) via the mcp__codex__codex tool for automated review of proposed optimizations.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and analyzes untrusted data from user interaction logs without sanitization or strict boundary enforcement. \n
Ingestion points: Raw user prompts and arguments are read from the .aris/meta/events.jsonl file. \n
Boundary markers: None are employed when interpolating log entries into the analysis and review prompts. \n
Capability inventory: The skill has access to file system modification tools (Edit, Write, Bash) and external model invocation tools. \n
Sanitization: No validation or filtering is performed on log content before it is processed by the logic engine.

meta-optimize