/impl-flow:allRun the full impl-flow pipeline (design -> plan -> implement) in one invocation, with a single checkpoint after the plan is drafted before implementation begins
/impl-flow:designInteractively interview the user to nail down requirements for a new implementation, record them as a use-case document at the root of the task-set directory, and harden it through an adversarial review by an independent agent before feeding into the plan phase (Phase 1 of impl-flow: design -> plan
/impl-flow:diagnosePin down a bug or change request against an existing use-case document, find the root cause, revise the affected use cases, and hunt for what a partial fix would miss through a three-layer omission review (Phase 1 of impl-flow's fix mode: diagnose -> plan -> implement). Not a standalone command — on
/impl-flow:fixFix mode — diagnose a bug or revision against an existing use-case document, revise the affected use cases, plan the fix, and implement it, with a single checkpoint after the plans are drafted. Lighter than /impl-flow:all, and aimed at not missing anything a partial fix would leave behind.
/impl-flow:implementExecute the pending implementation plans in a task-set directory. The main session acts as leader and dispatches each plan to an implementation subagent sized to that plan's declared complexity, running independent plans in parallel within a single branch (Phase 3 of impl-flow: design -> plan -> imp
/impl-flow:planTurn a design (or direct requirements) into commit-sized implementation plan files, each self-contained enough to run in a brand-new Claude Code session (Phase 2 of impl-flow: design -> plan -> implement). Not a standalone command — only invoked via /impl-flow:spec or /impl-flow:all.
/impl-flow:specRun design then plan back to back in one invocation (no implementation) — the two phases that are almost always used together