The Agent Skills Directory

[PROMPT_INJECTION]: The skill proactively addresses indirect prompt injection risks associated with its primary function of processing external documentation. It includes a 'Safety and quality standards' section that serves as a hard constraint for the agent.
Ingestion points: The skill processes untrusted backend routes, OpenAPI snippets, and API documentation provided by users (SKILL.md).
Boundary markers: The instructions explicitly state that ingested content is untrusted data, not instructions, and direct the agent to ignore any command-like text embedded in source materials (SKILL.md).
Capability inventory: The skill uses a Python script (scripts/scaffold_opencollection.py) to write generated requests to the local filesystem.
Sanitization: The scaffolding script implements slugify to sanitize filenames against path traversal and a quote function to ensure YAML values are safely escaped.
[COMMAND_EXECUTION]: The script scripts/test_scaffold_opencollection.py contains calls to subprocess.run(). This is used exclusively for unit and integration testing of the skill's own scaffolding script and is restricted to the local environment.
[CREDENTIALS_UNSAFE]: The skill instructions and documentation checklists emphasize secret management, mandating the use of placeholders like {{token}} and {{apiKey}} to prevent the accidental exposure of real credentials in generated files.

bruno-collection-generator