The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run within its test suite (scripts/tests/cli.py, scripts/tests/test_scaffold_skill.py) and starter templates to verify functionality. This is a standard and necessary behavior for a development framework that generates and tests scripts. To mitigate risks, the skill includes a dedicated security policy in references/exec-command-configuration.md which instructs the agent to use safe shell syntax, bounded outputs, and explicit working directories when executing commands.
[EXTERNAL_DOWNLOADS]: The skill references an external registry for skill reuse in references/registry.md. It points to github.com/srinitude/skills for canonical skill definitions. These references target well-known and trusted domains associated with the skill's vendor and are used to inform the agent's workflow rather than for automated remote code execution.
[INDIRECT_PROMPT_INJECTION]: As a tool that generates new skills from user-provided prompts, the skill naturally possesses an ingestion surface for untrusted data.
Ingestion points: User-supplied prompts for the new command and description flags in scripts/scaffold_skill.py.
Boundary markers: The references/generation-contract.md and references/writing-rules.md files provide strict guardrails for the generated output.
Capability inventory: The skill can write files, create directories, and run validation scripts using subprocess.
Sanitization: Multiple automated scripts (scripts/lint_writing.py, scripts/check_code_rules.py, scripts/check_placeholders.py) are used to validate the generated skill, ensuring it follows safety rules and does not contain unedited boilerplate or overly complex code before the build is marked as complete.

skill-factory