cross-agent-synthesizer·analysis·active
Cross-Agent Synthesizer
Consumes outputs from TARA, EVAN, LEO, and SARA in a single pass. Reconciles competing signals and surfaces the few that leadership should act on.
Cross-Agent Synthesizer
Consumes outputs from TARA, EVAN, LEO, and SARA in a single pass. Reconciles competing signals and surfaces the few that leadership should act on. When this skill fires
·An ARIA report is generated at sprint close ·An engineer asks:"what's the quality state of the release candidate?"
·A leadership-driven trigger (release gate, board prep, customer escalation)
Inputs
·TARA: scenario coverage, ambiguity flags from the requirement parser ·EVAN: test counts, pass/fail rates, stable vs. flaky breakdown ·LEO: bend points, soak findings, capacity headroom ·SARA: open findings by severity, time-to-remediate trend Reconciliation
→Different agents disagree. EVAN may say tests pass; LEO may say latency is degrading; SARA may say a new finding is open. ARIA's job is to weight these into one verdict, not to dump four separate reports. Weighting rules
·Severity dominates — a Critical SARA finding outweighs all green from elsewhere ·Trend dominates absolute — a 90% pass rate that was 95% last sprint is a worse signal than 88% steady
·Coverage gates the rest — if TARA's coverage report shows < 70% of acceptance criteria have scenarios, the rest of the metrics are low-confidence Conflict examples
·EVAN says all tests pass, but LEO says p99 is degrading → ARIA flags as"tests insufficient: latency regression not covered by EVAN's suite" ·SARA says no Critical findings, but the OWASP probe library has been disabled → ARIA flags as"security signal degraded: scope of probes reduced this sprint" What gets surfaced
ARIA's output is not a comprehensive dump. It's the 3–5 things leadership should know, ranked by: ·Action required (yes / no)
·Owner implied (specific team)
·Effort to act (hours / days / weeks)
·Risk if ignored (incident probability × blast radius)
Everything else lives in the deep-dive section, available on demand.
What it produces
A structured synthesis with:
·A one-line verdict "Quality is steady; one Critical security finding requires action this week")
·3–5 prioritized signals
·A trend annotation per signal (improving / stable / degrading)
·Confidence (high / medium / low) on the verdict
→The verdict is what gets read in the Monday standup. Everything else is what gets read by people who want to drill in.
Why ARIA, not a dashboard A dashboard with 40 metrics is exactly the same problem as four separate agent reports. ARIA's value is that itpicks — it makes the call that 37 of those metrics don't matter this sprint and 3 of them do, and it explains why.