القائمة
Use when a relay run needs its review anchor, scored rubric, or dispatch prompt — always before relay-dispatch, whether AC are explicit, vague, or missing.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Dispatch implementation tasks via worktree isolation. Use when delegating work to an executor, running background dispatches, or parallelizing independent tasks.
Use when completed relay executor work needs independent review against frozen Done Criteria — internal pre-publication rounds or post-publication PR rounds.
Drive already-planned relay leaves through fan-out, review, serial merge, crash recovery, and fleet status.
Interactive setup for relay routes. Use when the user asks to set up relay, configure company/personal relay routing, enable OpenCode or Pi, resolve model names, add provider/model routes, check advisory-review routing, or run relay-config doctor/check.
Use when a GitHub issue, sprint item, task description, or natural-language handoff should be implemented through autonomous executor dispatch; stops at ready_to_merge — merge only on explicit request.
Merge a reviewed PR, clean up worktree/branch, and close GitHub issues. Use after relay-review returns LGTM.
| name | relay-plan |
| argument-hint | [task, issue, or ready handoff] |
| description | Use when a relay run needs its review anchor, scored rubric, or dispatch prompt — always before relay-dispatch, whether AC are explicit, vague, or missing. |
| compatibility | Requires gh CLI. |
| metadata | {"related-skills":"relay, relay-ready, relay-dispatch, relay-review, dev-backlog","keywords":"계획, 루브릭, planning, rubric, dispatch prompt"} |
RELAY_SKILL_ROOT defaults to skills.AGENTS.md, CLAUDE.md, CHARTER.md, spec/capabilities.md, active sprint notes), optional /tmp/done-criteria-<N>.md, /tmp/dispatch-<N>.md, and /tmp/rubric-<N>.yaml.${RELAY_SKILL_ROOT:-skills}/relay-dispatch/scripts/reliability-report.js, ${RELAY_SKILL_ROOT:-skills}/relay-plan/scripts/probe-executor-env.js, ${RELAY_SKILL_ROOT:-skills}/relay-plan/scripts/persist-done-criteria.js, ${RELAY_SKILL_ROOT:-skills}/relay-dispatch/scripts/dispatch.js.relay-readyrelay-dispatchrelay-reviewrelay-mergerelay-plan emits handoff artifacts only; relay or an operator runs relay-dispatch.
Read the normalized task source (try in order, use first that succeeds):
~/.relay/requests/<repo-slug>/<request-id>/relay-ready/<leaf-id>.mdbacklog/tasks/{PREFIX}-{N} - {Title}.mdgh issue view <N>If relay-ready produced a handoff brief, treat it as the source of truth instead of re-reading the raw request.
For a shaped leaf, consume that persisted relay-ready/<leaf-id>.md plus its frozen Done Criteria path; the
raw request is historical context only and must not be silently reinterpreted into a different task.
If the relay-ready anchor is incomplete, surface the ambiguity or persist planner-authored Done Criteria under Step 7.
node "${RELAY_SKILL_ROOT:-skills}/relay-dispatch/scripts/reliability-report.js" --repo . --json
node "${RELAY_SKILL_ROOT:-skills}/relay-plan/scripts/probe-executor-env.js" . --project-only --json
Read historical relay signal, repo-local quality signal, and task-relevant local harness context as weak inputs only. They inform wording, prerequisites, commands, and where to look; they do not gate dispatch or override the task. Field meanings and authority hierarchy: references/signals.md.
Keep explicit AC, inferred Done Criteria, relay-ready handoff, project harness context, repo signal, historical signal, optional subsystem scout notes, and task risk as separate evidence channels until the review anchor is written.
Identify the evaluation source model:
If AC are missing, vague, or incomplete, write observable Done Criteria first. Treat explicit AC as high-priority evidence, not the only source. Before freezing, run the pre-flight ambiguity audit. If the final review anchor is planner-authored or differs from the task source, persist it in step 7.
Synthesize task intent, explicit AC when present, repo signals, and task risk into a scored rubric — the recovered Done Criteria are the anchor, not the issue text alone. Use the guided interview in references/rubric-design-guide.md to synthesize factors from them. Minimal shape:
rubric:
prerequisites:
- command: "npm test"
target: "exit 0"
factors:
- name: Observable task contract
tier: contract
type: automated
command: "<task-specific command>"
target: "<observable pass condition>"
weight: required
Size the rubric by task risk and ambiguity, not raw issue AC count. Tier classification, type, weight, setup/baseline, criteria, scoring_guide, size guidance, and optional tdd_anchor / tdd_runner: references/rubric-design-guide.md.
Quick gate before handoff: prerequisites hold repo-wide hygiene only; factors stay substantive; task-size minimums are satisfied; at least one automated check exists; evaluated factors have low/mid/high scoring_guide; criteria and targets are concrete. Full checklist: references/rubric-validation.md.
Before persisting, apply references/rubric-simplification.md: rewrite HOW into observable WHAT, merge overlaps, remove unsupported defensive clauses, and verify weights.
Persist only when planning writes the final Done Criteria, or expands, rejects, or narrows issue-body AC. This includes AC-missing inputs, user-provided descriptions, and any case where planning changes the issue-body AC.
Before Step 7, the orchestrator/planner must allocate the RUN_ID that dispatch will later reuse. Use the same valid run id in this persistence command and in Step 8's relay-dispatch --run-id "$RUN_ID" handoff; if no Done Criteria persistence is needed, dispatch may allocate the run id itself.
node "${RELAY_SKILL_ROOT:-skills}/relay-plan/scripts/persist-done-criteria.js" --repo . \
--run-id "$RUN_ID" --file /tmp/done-criteria-<N>.md --json
Skip this step when the issue or relay-ready handoff already provides the final Done Criteria without planner changes.
Write the dispatch prompt and rubric YAML to temp files. The prompt uses ../relay/references/prompt-template.md and appends Setup, optional Working Guidance, Scoring Rubric, Iteration Protocol, and Score Log sections. Full iteration protocol and Score Log format: references/iteration-protocol.md.
Return a handoff summary with dispatch prompt path, rubric YAML path, Done Criteria anchor path when persisted, review assurance level, and the recommended relay-dispatch command. Use --review-assurance hardened only when task/rubric risk requires stronger verification; never derive it from executor or reviewer identity.
If the recommended executor or reviewer route/model cannot resolve, point the operator to relay-config to register the route or set the default before dispatch.
When Step 7 persisted Done Criteria, the dispatch handoff must preserve both anchors:
node "${RELAY_SKILL_ROOT:-skills}/relay-dispatch/scripts/dispatch.js" . \
--run-id "$RUN_ID" --prompt-file /tmp/dispatch-<N>.md --rubric-file /tmp/rubric-<N>.yaml \
--done-criteria-file <done-criteria-path>
Do not run relay-dispatch from relay-plan.
Use add-ons only when task evidence earns them; do not copy reference checklists into the prompt.
scripts/match-template.js can suggest a scaffold; never auto-apply.task_profile or working guidance: references/task-profile.md and references/guidance-packs.md.references/rubric-domain-axes.md.references/rubric-trust-model.md and references/rubric-fail-closed-patterns.md.references/rubric-patterns.md.references/subsystem-scout.md; skip for S/M tasks with clear scope.references/rubric-stress-test.md; ambiguity or risk can opt any size into stress-test.