The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using git, gh (GitHub CLI), bash, jq, and grep. These are used to search for existing skills, parse API responses, modify configuration files, and manage the Pull Request lifecycle. These operations are restricted to the local environment and intended for repository maintenance.
[DYNAMIC_EXECUTION]: The skill generates new SKILL.md instruction files and updates aeon.yml configuration from natural language input. It uses a scripts/dry-run.sh script to structurally validate the generated content before it is finalized. This is a core feature for automated skill development and includes checks to prevent the generation of broken or malicious code.
[PERSISTENCE]: The skill modifies the aeon.yml file to register new skill schedules (cron jobs). While this creates recurring automated tasks, the changes are submitted via a Pull Request (gh pr create) and are disabled by default (enabled: false), requiring manual operator intervention and review to become active.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill accepts untrusted natural language input via the ${var} parameter in the SKILL.md file.
Boundary markers: The skill lacks explicit prompt delimiters to separate user input from its internal generation logic, creating a surface where a complex prompt could attempt to trick the generator.
Capability inventory: The skill has extensive file system, network, and repository management capabilities.
Sanitization: To mitigate injection risks, the skill employs a scoring system (1-5), regex-based post-write validation to remove placeholders or 'TODO' markers, and a structural dry-run gate that prevents the creation of PRs for skills that fail security or quality checks.

create-skill