菜单
Use to reconcile drift across related Final, unimplemented RDRs before any implement. Runs Stage 7.1 whole-set critique and pairwise contradiction scan. Trigger for cluster reconcile, $rdr-cluster-reconcile, or /rdr-cluster-reconcile.
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
Use to run one pre-lock lens and resolve its findings in the same pass. Lenses: grounding, 3amigo, critique, repeatability, cove. Trigger for pre-lock review, $rdr-prelock, or /rdr-prelock.
Use to close open spikes and round-disturbed assumptions before lock. Runs Stage 6 and routes back if evidence refutes the draft. Trigger for reconcile RDR, $rdr-reconcile, or /rdr-reconcile.
Use to find the current stage and next command for an RDR. Read-only navigator over status, evidence, and artifacts. Trigger for where is RDR, next RDR step, $rdr-status, or /rdr-status.
Use to implement a locked Final RDR into working, spec-verified code. Runs Stage 8 via the launch prompt and is re-entrant from status.md. Trigger for implement RDR, $rdr-implement, or /rdr-implement.
Use to verify an RDR setup. Read-only health check for the seam, five-var contract, engine layout, evidence root, and skill links. Trigger for diagnose RDR, check RDR setup, $rdr-doctor, or /rdr-doctor.
Use to verify critical RDR assumptions against source, spikes, prior art, and reuse audits. Runs Stage 4 before pre-lock review. Trigger for resolve assumptions, $rdr-resolve, or /rdr-resolve.
| name | rdr-cluster-reconcile |
| metadata | {"argument-hint":"<cluster-name | NNNN NNNN [NNNN…]> [--commit | --no-commit]"} |
| description | Use to reconcile drift across related Final, unimplemented RDRs before any implement. Runs Stage 7.1 whole-set critique and pairwise contradiction scan. Trigger for cluster reconcile, $rdr-cluster-reconcile, or /rdr-cluster-reconcile. |
Per-RDR stages check one RDR at a time; cross-RDR consistency is a set-level concern no per-RDR gate covers. This stage reconciles a cluster of related RDRs — all Final-and-unimplemented — once, before any of them implements. Per cluster, not per RDR. Most RDRs skip it (it applies to foundational / cross-RDR producers).
Codex: $rdr-cluster-reconcile <cluster-name | NNNN NNNN [NNNN ...]>
Claude: /rdr-cluster-reconcile <cluster-name | NNNN NNNN [NNNN ...]>
Takes a cluster (a name you give the set, or the list of RDR numbers in it) — not
a single RDR. If only one RDR is Final-and-unimplemented, there is no cluster;
say so and point at /rdr-implement NNNN.
rdr-common.md; run §seam-bind to bind $RDR_ENV,
$RDR_RESOURCES, and $RDR_RECORDS (the consumer's RDR directory, exported by the
marker) plus the output base <EVIDENCE_DIR>/cluster-reconcile/<cluster>/.{RDR_RECORDS};
form the peer pairs to compare.07.1-cluster-reconcile.md owns the
mechanics — read it for the cluster-membership and re-entry-scope calls):
2-critique.md
run against the set; writes critique-set.md.pairwise.md
run per pair; writes pairwise-<A>-<B>.md.
Delegate the heavy reads (several RDRs, the pairwise runs) to a sub-agent that
returns verdict + evidence pointer (§return-packet).07.1-cluster-reconcile.md)chore(rdr): cli/NNNN cluster-reconcile <cluster>); the doc/lock commit lands at finalize, not here.Next: /rdr-implement NNNN for each cluster member./rdr-propose /
/rdr-refine / /rdr-resolve per the scope), re-lock via /rdr-finalize NNNN,
then implement./rdr-status NNNN to re-orient any member.