| name | xc881-requirement-analysis |
| description | Primary owner for requirements, scope, acceptance criteria, PRDs, and issue/PR triage. Separate explicit/inferred needs, persist task queues when warranted, and delegate only blocking research or reads. Implementation detail alone is not requirement intent. |
| when_to_use | Use for scope/AC/tracker deliverables. Persist only when the gate passes; delegate bounded uncertainty, then resume the same owner. |
| display_name | xc881需求分析 |
| version | 2.10.0 |
| category | requirements |
| tags | ["intent-routing","requirements","prd","mvp","acceptance-criteria","issue-triage","tracker","xc881"] |
| aliases | ["xc881-requirement-analysis","xc881需求分析","需求分析","xc881分诊","xc881拆issue"] |
xc881 Requirement
Intent-routed requirement owner. Direct calls select it; implicit use requires
scope/AC/tracker as the primary outcome. Incidental implementation uncertainty
stays with the base/current owner. Run only the needed branch. No code.
Goal:
request → spec → ER/IR → scope → AC → risk → handoff
Reference files live under ./references/.
Reference Index:
./references/english-spec-first-policy.md: rough/mixed request.
./references/implicit-requirement-inference.md: hidden needs.
./references/existing-project-feature-analysis.md: existing change.
./references/issue-tracker-triage.md: issue/PR/PRD tracker state, Agent Brief, out-of-scope, and tracker ops.
./references/requirement-process-artifact.md: artifact gate, req-process queue, resume/update protocol.
./references/requirement-output-contract.md: output.
./references/requirement-quality-rubric.md: strict check.
When opening reference files, always use the exact relative path shown above.
Do not assume reference files are located beside SKILL.md.
Steps:
- Confirm requirement/triage intent from direct invocation, wording, or delegated context.
- Spec:
Goal / Inputs / Constraints / Output / Notes.
- Classify: new / existing / review.
- Resolve an explicit or unambiguous conversational artifact first. Run bounded
candidate-path search only when persistence or implementation handoff is likely;
clearly chat-only work does not scan req-process. Reuse one semantic match, ask
once for multiple matches, or apply the gate when none match.
- If technical uncertainty blocks scope or AC judgment, use bounded
$xc881-solution-research, then continue without user reconfirmation.
- If project code/test/config must be inspected, apply the task+project complexity gate: perform a simple exact/local/low-risk read here, or delegate a complex/tracked
read to $xc881-coding-skills, then resume from evidence.
- Produce
ER/IR, scope, FR/NFR/AC, and P0-P3 risks; one blocking question max.
- If the artifact gate passes, create dependency-aware vertical queue items for
Coding. Otherwise keep a compact self-contained chat contract. Reused artifacts
preserve history, append IDs after the tail, and reopen from the first new item.
Persist verification plans and queue state, never results/logs.
- Keep the full structured analysis in req-process when persisted, then report a compact human-readable summary. Chat-only work must remain self-contained.
- Validate, then hand every write and selected complex/tracked read to Coding with exact artifact/current/next/blockers. A queued implementation continues across items; after context compaction it resumes the same owner/current automatically. Active ownership returns to this requirement run after bounded delegation, and no second invocation or per-item “continue” is needed unless the user requests stepwise control or a real blocker occurs.
User-facing output:
需求结论:
这次会做:
完成标准:
需要注意: # only when relevant
需要确认: # only when blocked; include recommendation + fallback
下一步:
开发记录:
Rules: route only on requirement/triage intent; no code; inferred ≠ confirmed; AC must be testable; persisted documents retain Spec/ER/IR/Scope/FR/NFR/AC/Risk while chat translates decisions into plain language; omit irrelevant empty headings; persisted progress is authoritative over chat; use $xc881-solution-research only for bounded technical uncertainty, and return its result to the same requirement run.