The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill collects untrusted input from the user (descriptions, triggers, and names) and injects it into new SKILL.md and README.md files.
Ingestion points: Phase 1 'Brainstorming' session via user prompts.
Boundary markers: No explicit delimiters or instructions are used to separate user-provided content from template instructions in the generated output.
Capability inventory: The tool executes file writes, directory creation, and symlink creation across the file system.
Sanitization: User input is not sanitized before being used in shell commands (sed) or written to generated files, which may lead to malformed instructions or injection into the downstream agent's context.
[COMMAND_EXECUTION]: The skill uses shell scripts to perform environment discovery (checking for gh and .claude directories) and setup tasks. It relies on standard utilities like sed, mkdir, and ln to manage the skill lifecycle.
[DYNAMIC_EXECUTION]: The skill creates and modifies executable instructions (SKILL.md) and helper scripts at runtime. The scripts/init_skill.py utility specifically generates Python scripts with executable permissions (chmod 0o755) based on templates.
[PERSISTENCE]: The skill provides a feature to 'globally install' generated skills by creating symlinks in persistent configuration directories like ~/.copilot/skills/ and ~/.claude/skills/. This ensures the generated capabilities remain available to the agent in future sessions.

skill-creator