The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data during Sendbird migrations, which presents a potential injection surface.
Ingestion points: The sendbird-data-migration.md file describes exporting data from Sendbird into migration.jsonl files for processing.
Boundary markers: The skill explicitly requires user confirmation before starting any data migration ("Ask the user first
never start a data migration unsolicited").
Capability inventory: The skill uses the getstream CLI, curl for uploads, and file system read/write operations.
Sanitization: The instructions mandate validating the generated jsonl file against a JSON Schema before import to ensure data integrity and catch malformed content.
[DATA_EXFILTRATION]: The skill implements exemplary practices for secret management to prevent exposure.
RULES.md strictly prohibits reading or editing .env files directly in the chat interface to prevent secrets from leaking into conversation logs.
It requires using the getstream env command for credential management and mandates verifying .gitignore entries before writing any sensitive environment variables.
[COMMAND_EXECUTION]: Shell command execution is limited and follows a "confirm-before-write" policy.
The SKILL.md frontmatter restricts the Bash tool usage to getstream * subcommands only.
RULES.md requires the agent to describe and obtain explicit user confirmation before any operation that creates, modifies, or deletes data.

stream