菜单
Inventory, investigate, cluster, and provisionally classify KlumAST GitHub issues for release planning. Use for global issue sweeps, duplicate discovery, 4.0 scope analysis, or refreshing the persistent curation indexes without mutating GitHub issues.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Build a domain-first KlumAST Schema. Use when a Domain API Developer and Schema Developer want to model a domain independently of downstream targets, decide between Layer 3 and direct-schema, define DSL Object boundaries, and prove the resulting completed model with an API-only client and executable test.
Build a KlumAST authoring Schema for an authoritative external contract. Use when an adopter is replacing or simplifying Helm values, another configuration document, or a target-specific API contract while keeping that contract authoritative.
Advise on applicable KlumAST features in an existing Schema or configured model. Use when an adopter wants a ranked, evidence-based KlumAST improvement review, an explanation of supported 4.x features that fit their model, or selected, validated model improvements.
Author a configured KlumAST model. Use when a Model Writer wants to turn a representative configuration into a KlumAST model, add validation, construct it through the generated DSL, or create an executable model test.
Start or adapt a Gradle project for KlumAST. Use when an adopter wants to scaffold a KlumAST project, add KlumAST to an existing Gradle build, choose domain-first or target-contract modeling, choose Layer 3 or direct-schema structure, or verify a first schema build.
Maintain a compact, evidence-backed horizon of user-selectable repository work; create and reconcile user-visible tasks without confusing completed execution with delivered repository state.
| name | klum-curate-issues |
| description | Inventory, investigate, cluster, and provisionally classify KlumAST GitHub issues for release planning. Use for global issue sweeps, duplicate discovery, 4.0 scope analysis, or refreshing the persistent curation indexes without mutating GitHub issues. |
Build an evidence-backed issue inventory before interviewing the maintainer or changing GitHub.
AGENTS.md, CONTEXT.md, docs/agents/issue-tracker.md, docs/agents/triage-labels.md, docs/agents/feature-triage.md, and docs/agents/domain.md.README.md, CHANGES.md, wiki/Builder-First-Migration.md, relevant ADRs and implementation notes, and docs/implementation/issue-curation/architecture-map.md if present.wiki/Roadmap.md as historical evidence unless current decisions confirm it.Use the repository's configured GitHub workflow to read every open issue, including comments, labels, milestone, dates, and linked work. Do not edit, label, close, or retarget issues in this skill.
For each issue record:
4.0 must, 4.0 nice, 4.1, later 4.x, 5.0, completed, obsolete, duplicate, or maintainer decisionUse the feature-triage classification as priority evidence and derive minimum scope from the primary use case. Do not block classification when evidence is unknown, revisit already-triaged enhancements solely to fill the fields, or let a secondary angle silently broaden the request. When an existing enhancement is being reconsidered for another reason, add the record from available evidence. A complexity-free mini-feature may instead carry a concise quick-win rationale.
Cluster by underlying domain problem before deep source exploration. Inspect one code path per cluster where possible and record where redundancy and prior-rejection checks were performed.
Create or refresh only:
docs/implementation/issue-curation/issue-index.mddocs/implementation/issue-curation/duplicates.mddocs/implementation/issue-curation/release-4.0.mddocs/implementation/issue-curation/decision-log.mdDistinguish evidence, inference, and unresolved maintainer decisions. Preserve confirmed earlier decisions unless new evidence explicitly contradicts them. Finish only when every open issue appears exactly once in the index, every cluster is cross-linked, and uncertain decisions have targeted interview questions.