orca-workflows

Build a multi-part feature or migration as a supervised Orca task DAG with decision gates — plan, gate the spec for user approval, run independent components in parallel waves, then integrate and verify. Use this skill whenever the user asks to implement, build, or develop something spanning several modules or services with controlled checkpoints — "build feature X", "xây tính năng", cross-service migrations, "split this across agents", or any implementation where they want to approve the approach before code lands. Not for sequential bug-fix chains, which stay with a single agent.

Entry point and router for any Orca multi-agent request. Use this skill whenever a request involves coordinating several agents, supervising or monitoring agent work, waiting for results, task DAGs or decision gates, handing work off to another agent or worktree, checking worker status or mail, or anything Orca — even if the user only says "relay this", "spawn a team", "have another agent look at it", "giao việc cho agent khác", "không cần theo dõi đâu", "nhắc anh khi xong", or asks for one of this plugin's workflows without naming Orca. The router classifies the request, picks supervised orchestration vs. fire-and-forget handoff vs. plain terminal work, then loads the version-matched guide from the live binary before any command runs.

Review one diff, branch, or PR with several supervised Orca reviewers in parallel, each with a distinct lens, then merge and dedupe findings. Use this skill whenever the user asks for a code review, PR review, security review, pre-merge check, or a second opinion on changes — "review giúp anh", "check this diff", "look for issues before we merge" — especially on changes large or risky enough that a single pass would stay shallow.

Fan out a broad research, analysis, or investigation question across parallel supervised Orca workers, then synthesize their findings. Use this skill whenever the user wants something researched, compared, investigated, or explored along several independent angles at once — "research X", "phân tích giúp anh", "compare our options", codebase deep-dives too large for one context, market or technology surveys, and multi-part questions — even if they don't mention agents or Orca. Workers each write findings to a report file; the coordinator synthesizes from those files instead of passing large outputs through messages. Not for root-causing one specific failure, which stays depth-first with a single agent.