`sdd_clarify` runs an 8-dimension ambiguity analysis and blocks until the score clears its threshold — 70 in guided mode, 50 in expert — which is the stage that stops an agent building something nobody asked for. For a repo with no specs at all, `sdd_reverse_engineer` reads the code and writes nothing; files appear only when you run `sdd_bootstrap`.
A single Go binary that adds five systems to any MCP-compatible AI tool: persistent memory, an adaptive change pipeline, a full project pipeline, codebase bootstrap, and standalone review tools.
- Memory across sessions — `mem_save`, `mem_search`, `mem_context` and `mem_timeline` over SQLite with FTS5 full-text search
- A knowledge graph: `mem_relate` connects observations with typed edges such as `depends_on`, `caused_by`, `implements` and `supersedes`
- A change pipeline (5 tools) whose `sdd_context_check` stage scans existing specs, completed changes, memory and convention files like CLAUDE.md for conflicts before work starts
- A project pipeline (10 tools) running init → principles → charter → specify → business-rules → clarify → design → tasks → validate
- Bootstrap for code that has no specs: `sdd_reverse_engineer` produces a read-only evidence report, `sdd_bootstrap` writes the missing requirements, business rules and design files
- Four standalone tools that need no active pipeline — `sdd_explore`, `sdd_suggest_context`, `sdd_review` and `sdd_audit`
Install the binary — `brew install HendryAvila/hoofy/hoofy`, the install script, or `go install github.com/HendryAvila/Hoofy/cmd/hoofy@latest` (Go 1.25+). Then register it in one command: `claude mcp add --scope user hoofy hoofy serve`. One binary, no external runtime dependencies.
One command — brew install HendryAvila/hoofy/hoofy