AgentSkillsCN

requirement-elicitation

在需求撰写前的证据收集工作流程。仅当Codex需从各类文档与对话中收集并梳理需求输入时使用;切勿用于正式的需求基线整合——这一环节应归属于需求定义阶段。

SKILL.md
--- frontmatter
name: requirement-elicitation
description: Evidence gathering workflow before requirement writing. Use only when Codex must collect and structure requirement inputs from artifacts and conversations; do not use for formal baseline synthesis, which belongs to requirements-definition.

Requirement Elicitation

Trigger Boundary

  • Use when requirements are not yet explicit and evidence must be gathered.
  • Do not use to finalize canonical requirement wording; use requirements-definition.
  • Do not use to rank release scope; use requirement-prioritization.

Goal

Collect high-confidence requirement inputs with provenance and conflict visibility.

Shared Requirements Contract (Canonical)

  • Use ../requirements-definition/references/requirements-governance-contract.md as the single schema and gate source.
  • Track requirements workflow artifacts with RQM-* IDs.
  • Run machine validation: python3 ../requirements-definition/scripts/validate_requirements_contract.py --manifest <path/to/manifest.json>.

Inputs

  • Stakeholder list and decision map
  • Existing docs, tickets, incident records, and analytics
  • Initial hypotheses and known constraint areas

Outputs

  • Requirement candidates with confidence and source IDs
  • Conflict list and dependency findings
  • Prioritized clarification questions

Workflow

  1. Define evidence targets by uncertainty and business impact.
  2. Gather signals from interviews, documents, and system artifacts.
  3. Register findings with INT-*, UR-*, or EVD-* IDs.
  4. Classify findings as requirement, constraint, assumption, or unknown.
  5. Deduplicate terms and normalize vocabulary before handoff.

Quality Gates

  • Every candidate has an auditable source ID.
  • Contradictions are documented, not merged silently.
  • Unknowns are prioritized by delivery risk.
  • Personal data handling controls are defined before circulation.

Failure Handling

  • Re-run collection when findings rely on single weak sources.
  • Stop when lawful basis or retention policy is undefined for collected data.