Skills library for your AI Code that trades interviewing for building: fast opinionated specs and TDDs, then parallel hordes of cheap agents driving to a running prototype
Skills 15 in the Claude listing
Assumption ledger Use when logging decisions hordev made without asking the user, or reviewing open assumptions for falsification Debugging in a horde Use when a bug appears in code written by parallel agents who never saw each other's work Decomposing for hordes Use when turning a TDD into independent tasks that can run in parallel without shared file writes or dependencies Dispatching hordes Use when a decomposed task list with file ownership is ready and agents need launching, or when a dispatched agent times out, returns a plan instead of code, fails its tests, or reports BLOCKED Horde qa Use after reconciling-horde-output leaves a tree that builds and passes its tests, before telling the user anything is done, or whenever you are about to write "tests pass", "working", or "complete" a Horde status Use when asked where a hordev run stands - "where are we", "what's the horde doing", "status", "what's left", "how's the run going", "what did you decide" - or before handing a run back to the user Improving hordev Use when a run has ended badly or surprisingly, when the same failure has appeared more than once, when .hordev/run-log.md has grown since the last reflection, or when asked to make hordev itself bett Isolating horde workspaces Use before dispatching-hordes launches its first wave, before racing-prototypes creates candidates, and when tearing down worktrees after horde-qa passes Prototype first Use when deciding what to build first, when about to add config, flags, an abstraction layer, error handling for undescribed cases, or tests beyond the TDD, or when deciding whether the prototype is d Racing prototypes Use when two or more approaches are plausible and argument cannot decide between them Rapid spec Use when a request needs a spec before anything can be built, when requirements are vague and the instinct is to interview the user, or at the start of any hordev run Reconciling horde output Use when horde agents have returned and their files have not yet been built or tested together - the first pass that makes parallel output compile, agree at the seams, and match the TDD's names Using hordev Use when starting any conversation - establishes the hordev posture, when to build instead of ask, and which skill handles which stage of the run Writing hordev skills Use when creating a new skill, editing an existing one, or verifying a skill triggers correctly and works standalone Writing tdds Use when a spec exists in .hordev/specs/ and no TDD exists for it yet, or when horde-qa sends a TDD back for a missed requirement or a weak test