SuperAPI AI Directory

multi-role-quality-synthesis

Testing & quality

View repository

Updated 10-16 06:50

About this skill

Multi-Role Quality Synthesis (Chinese Version)

When to Use

  • One or more role reports already exist for the same quality stage, and a unified, traceable stage synthesis must be formed.
  • Multiple roles raise duplicate, complementary or conflicting quality findings, and sources must be preserved and disagreements explicitly handled.
  • Project delivery constraints and quality facts appear at the same time, and scheduling or stakeholder requirements must be prevented from overriding quality evidence.

Input

Receive one or more role reports. Each report should contain stage、source_role, facts, evidence, findings, risks, gaps, issues, actions and confidence levels, and should provide a unique source_id. Missing fields may only be recorded as input gaps and must not be fabricated.

Execution Flow

  1. Read and fully follow prompts/multi-role-quality-synthesis.md; this Prompt is a standalone execution specification.
  2. Verify that there is at least one report, that report sources are locatable, and that stage is consistent. Different stages must not be mixed into a single quality conclusion.
  3. Build a source registry, classify facts, quality findings and project constraints, and then merge.
  4. Merge only semantically equivalent findings and join all sources; keep conflicting findings separate and place them in the disagreement section.
  5. Output a traceable synthesis, then perform self-checks on sources, disagreements, severity levels, project boundaries and no newly added facts.

Core Constraints

  • Each synthesized finding cites at least one input source_id; equivalent findings retain all contributing sources and roles.
  • Do not use voting or unexplained averaging. When severity levels conflict, retain each source's level; when a synthesized level is needed, explain it according to the Prompt's deterministic rules.
  • P0/P1 raised by a minority source must not be hidden, downgraded or removed due to majority opinion.
  • Disagreements must be output explicitly; when there is no disagreement, also state "no disagreement".
  • Project schedules, capacity, dependencies, milestones and stakeholder requirements are separated from quality facts; project input cannot change defects, execution, evidence or quality status.
  • Do not generate any new facts, evidence, root causes, impacts, statuses or conclusions that none of the input reports provide.
  • This Skill does not depend on any specific role Skill. It only processes the role reports provided by the caller and does not load or invoke other role Skills.

Output Contract

Output in order:Stage and input coverage, quality facts, project constraints, synthesized findings, consensus, disagreements, blockers, information gaps, open questions, actions, confidence and limitations, source registry. Findings, consensus, disagreements, blockers and actions all retain sources.

Load on Demand

  • On every execution, read prompts/multi-role-quality-synthesis.md, which contains the complete input validation, merge order, severity level rules and output template.
  • Use evals/when evaluation or regression is needed; eval scenarios are not project facts and must not be brought into the real synthesis.

Pre-Delivery Self-Check

  • All inputs belong to the same stage, or cross-stage merging has been explicitly blocked
  • Each synthesized finding has one or more input sources
  • Equivalent findings retain all contributing sources, and conflicting findings have not been silently merged
  • P0/P1 minority opinions remain clearly visible; severity levels are not averaged without explanation
  • The disagreement section is always present; when there is no conflict, it explicitly states that there is no disagreement
  • Project constraints and quality facts are kept separate; PM/delivery input does not override quality facts
  • No facts, evidence, root causes, impacts, or conclusions unsupported by new input are added

Common Misconceptions

  • Do not force-merge findings with similar wording but different objects, conditions, or impacts.
  • Do not treat "the majority of roles agree" as a reason to remove minority high-risk opinions.
  • Do not treat deadlines, resource pressure, or status change requests as quality evidence.
  • Do not fabricate the perspective of an optional role that is absent; it should be marked as an input coverage gap and a confidence limitation.