The Agent Skills Directory

[SAFE]: The skill facilitates database migrations through structured Harness DBOPS tool calls (harness_get, harness_list, harness_execute). All operations are performed within the scoped environment of the Harness MCP server.
[COMMAND_EXECUTION]: The skill uses harness_execute to trigger database pipelines. This is the primary intended purpose of the skill and is gated by a mandatory user review process (Step 5: Present for Review) where users must explicitly choose 'Accept', 'Accept & Commit', or 'Deny'.
[DATA_EXPOSURE]: The skill accesses database metadata (schema names, table structures) to assist in generating changesets. This data is handled through standard Harness API resources (database_schema, database_snapshot_object) and is not exfiltrated to external third-party domains.
[PROMPT_INJECTION]: The skill includes instructions to ignore embedded instructions in user data (Step 3-B conflict checks), but these are defensive measures rather than malicious injections. There are no patterns suggesting attempts to bypass safety filters or extract system prompts.
[DYNAMIC_CONTEXT_INJECTION]: The skill contains a reference to resolving a conversation_id from Cursor chat transcripts. While this involves local file path inspection (~/.cursor/projects/...), it is a documented requirement for session mapping in the development environment and does not involve silent exfiltration or unauthorized command execution.

dbops-changeset