Skip to content

rfe-creator

A comprehensive Claude Code skill suite for the full lifecycle of Requests for Enhancement (RFEs) in the RHAIRFE Jira project. Covers creation from problem statements, rubric-based review with auto-revision, intelligent splitting of oversized RFEs, and submission to Jira. Also provides strategy document skills (RHAISTRAT) for refining approved RFEs into implementation strategies with adversarial multi-reviewer validation.

The plugin uses a shared artifact convention -- all skills read from and write to an artifacts/ directory with YAML frontmatter for structured metadata. Jira write operations use deterministic Python scripts rather than LLM tool-calling, while read operations support both Atlassian MCP and REST API fallback. A dependency on the assess-rfe plugin provides the scoring rubric, bootstrapped automatically on first use.

Plugin Details

Pipeline

rfe-creator pipeline

Dependencies

Skills

Skill Description Invocable
/rfe.create Generate new RFEs from problem statements
/rfe.review Score and improve RFEs with auto-revision
/rfe.split Decompose oversized RFEs into appropriately-scoped pieces
/rfe.submit Push RFEs to Jira
/rfe.speedrun Execute the full RFE pipeline end-to-end
/rfe.auto-fix Batch review, revise, and split operations
/strat.create Create strategy documents
/strat.refine Refine strategy documents
/strat.review Review strategy documents
/strat.prioritize Prioritize strategy items
/rfe-creator.update-deps Update vendored dependencies
/architecture-review Architecture review skill
/feasibility-review Feasibility review skill
/rfe-feasibility-review RFE feasibility review
/scope-review Scope review skill
/testability-review Testability review skill

Installation

/plugin install rfe-creator@opendatahub-skills

Architecture

Two skill families: RFE skills (rfe.) for the requirements pipeline and Strategy skills (strat.) for implementation planning. A speedrun skill orchestrates the full end-to-end flow by invoking other skills.

Review skills use a forked reviewer pattern -- independent sub-agents (feasibility, testability, scope, architecture) run in isolated contexts and produce separate assessments that are synthesized into a consolidated review. The rfe.review skill is the central orchestrator: it launches parallel waves of fetch, assess, feasibility, review, and revise agents, polling for completion via scripts/check_review_progress.py.

State persistence uses scripts/state.py for long-running operations across context compression boundaries, and scripts/frontmatter.py manages YAML frontmatter on all artifact files. The rfe.auto-fix skill wraps this into a pipeline state machine (scripts/pipeline_state.py) that handles batching, resumption, and multi-phase dispatch with launch_wave/wait-for-wave synchronization.

Architecture context is fetched from opendatahub-io/architecture-context into .context/architecture-context/ and used by review and strategy skills to ground assessments in actual platform components and APIs.