The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/install.sh modifies the ~/.codex/AGENTS.md file by appending rules from the skill. This is a global configuration file that determines agent behavior across all sessions, and modifying it represents a persistent change to the agent's core instructions.
[EXTERNAL_DOWNLOADS]: The installation guide recommends using npx skills add to fetch the skill from https://github.com/nangongwentian-fe/jay-skills. While this repository belongs to the author, the process involves downloading and preparing code from a remote source for execution.
[PROMPT_INJECTION]: The skill's primary function is to analyze task history for learning, which presents a risk of indirect prompt injection.
Ingestion points: The SKILL.md review workflow instructs the agent to analyze task logs and results to identify lessons.
Boundary markers: The instructions do not define delimiters or provide warnings to ignore malicious commands that might be embedded in the processed task data.
Capability inventory: The skill has the capability to update agent memory, modify project documentation, and create or update other skills using the $skill-creator tool.
Sanitization: There is no mechanism described to sanitize or validate extracted lessons before they are used to update system instructions or persistent documentation.

post-task-learning-review