Author and refresh roadmap concept explainer pages at `roadmap/concepts/<slug>.html` (R486) — interactive, intuition-first HTML background pages for dense or recurring roadmap concepts. Use when the user asks to "write an explainer", "generate a concept page for R24", "explain the nodeID concept as a page", "refresh the join-paths explainer", or any phrase about authoring or updating `roadmap/concepts/` pages. Two intents: generate (author a new page from an item ID or concept slug) and refresh (recompute only the derived regions of an existing page).
Produce a single copy-pasteable code block that prompts another agent to perform a focused, architecture-oriented code review of the current branch (or a user-specified target). Use when the user asks for a "reviewer prompt", "review prompt", a "prompt for another agent to review", or otherwise wants a hand-off prompt for a fresh reviewer. The emitted prompt instructs the reviewer to find improvement opportunities aligned with graphitron-rewrite's design principles, not to bug-hunt, rubber-stamp, or police style.
Produce a single copy-pasteable code block that prompts another agent to perform a workflow-gate review on a graphitron-rewrite roadmap item. Use when the user asks for a "spec review prompt", "review prompt for R<n>", "Spec → Ready handoff", "In Review → Done handoff", or otherwise wants a hand-off prompt for the workflow-gate review on a specific roadmap item. The emitted prompt encodes the relevant reviewer-rule and pre-fills recent commit history so the next reviewer can apply the rule without digging.
Manage graphitron-rewrite roadmap items — add new Backlog entries (auto-allocates an `R<n>` ID), transition status (Backlog→Spec→Ready→In Progress→In Review→Done), and regenerate the rolled-up README. Use when the user asks to "add a roadmap item", "move R24 to Ready", "flip R12 to In Review", "mark R7 done", or any phrase about roadmap state. Items are referred to by `R<n>` (lookup by id is fast); slug is the filename only. Encodes the workflow.md state machine and the reviewer-rule guard.
Maintain the graphitron capability catalog at `capabilities/<slug>.adoc` — add, edit, suggest, audit, dedupe, and remove capabilities. Use when the user asks to "add a capability for X", "list capabilities", "suggest missing capabilities", "audit the catalog", "remove the foo capability", "edit the pagination capability", or any phrase about graphitron's capability namespace. The catalog is the source of truth for the slug namespace `@capability(name:)` and `@exemplifies(capability:)` reference; every authored slug must surface on at least one coordinate, every directive value must resolve to an authored slug.
Investigative-journalist-style discovery skill that walks graphitron's model (sealed variants, records, directive resolvers, SDL types), asks Socratic questions about what each thing exists to deliver, and turns answers into capability catalog stubs and candidate `@capability(name:)` SDL tags. Use when the user asks to "interview the model", "discover capabilities", "walk the codebase for capabilities", "what capability does X serve", or otherwise wants conversational capability discovery rather than the heuristic walk `capability-catalog suggest` does. Sibling to capability-catalog: this skill discovers; that skill records.
Migrate one classification verdict into the R281 spec-by-example corpus and the code-generation-triggers documentation, retiring the legacy GraphitronSchemaBuilderTest enum row it subsumes. Use when grinding R281 slice 2/3, or when the user says "migrate the X verdict", "add a corpus example for X", "document the X classification", "render X from the corpus", or "retire the X enum case". Encodes the per-verdict loop: author fixture, validate dimensions, render the doc block, write the worked example, retire pure-verdict enum rows, verify coverage + drift + docs render.
Delete remote branches whose commits are all contained in the claude/graphitron-rewrite trunk. Use when cleaning up dead/merged Claude branches off the rewrite trunk.