Create one immutable, single-agent technical implementation plan from a fixed spec diff, limited to behavior still missing at the fixed head. Use when a feature, fix, migration, or refactor has approved spec changes and needs a sequential implementation plan before execution is divided or orchestrated.
Divide an explicit immutable implementation plan into context-sized TDD slices and operate a recoverable Herdr execution ledger through independent test-quality and final reviews. Use when a create-plan artifact is ready for isolated implementation, verification, integration, remediation, or interrupted execution recovery inside Herdr.
Bootstrap a specs/ folder with a language-agnostic specification suite for defining business rules, logic, and requirements. Conducts a grill-style interview to discover systems, bounded contexts, and terminology, then generates SPEC-OF-SPECS, README, ubiquitous language, design language, skeleton specs, and a progress tracker. Use when starting a new project, onboarding AI agents, creating a specification suite from scratch, or initializing specs/ for any project. Supports greenfield (generate directly) and brownfield (generate PLAN.md for agent-driven spec extraction from code).
Curate durable agent improvements from a human retrospective, current session, recent history, git activity, skills, docs, specs, and plans. Use when the user runs /curator, asks what should be learned from a session, wants compounding engineering recommendations, asks to steer recommendations with a retrospective, or asks whether to add, update, remove, rename, refactor, or consolidate skills or docs.
Interview the user relentlessly about a plan or design until shared understanding resolves every consequential branch, using project terminology and evidence. Use when stress-testing a plan, probing a domain model, resolving design decisions, or when the user says "grill me."
Launch and reliably coordinate Claude Code in Herdr by composing the base Herdr transport skill, handling readiness, explicit prompt submission, concurrent completion/blocking observation, and evidence-driven steering. Use when a caller selects Claude Code as a Herdr worker or reviewer.
Control herdr from inside it. Manage workspaces and tabs, split panes, spawn agents, read output, and wait for state changes — all via CLI commands that talk to the running herdr instance over a local unix socket. Use when running inside herdr (HERDR_ENV=1).
Audit skills in shared/skills/ for cross-agent frontmatter compatibility, flagging missing fields and offering to fix them. Use when skills have been added via npx or manually and need to be checked for compatibility across Claude Code, Codex, Pi, and Copilot CLI.