Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/gonzalezpazmonica/saviaWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/gonzalezpazmonica/savia/recommendation-tribunal-orchestrator)<a href="https://agentmods.dev/agents/gonzalezpazmonica/savia/recommendation-tribunal-orchestrator"><img src="https://agentmods.dev/badge/agents/gonzalezpazmonica/savia/recommendation-tribunal-orchestrator/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/gonzalezpazmonica/savia/recommendation-tribunal-orchestrator"><img src="https://agentmods.dev/badge/agents/gonzalezpazmonica/savia/recommendation-tribunal-orchestrator.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00043 | $0.01132 |
| Opus 5 | $0.00022 | $0.00566 |
| Sonnet 5 | $0.00009 | $0.00226 |
| Haiku 4.5 | $0.00004 | $0.00113 |
Grade A, and why
recommendation-tribunal-orchestrator scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
Copies of this mod
1 near-identical copy found in the catalogue:
- recommendation-tribunal-orchestrator — 100% identical, 0 lines differ
How it starts
The opening of the file, as written. The whole thing — 86 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Recommendation Tribunal Orchestrator — SPEC-125 Slice 1
You convene the 4-judge Recommendation Tribunal for conversational recommendation reliability. You orchestrate (do not judge); aggregate verdicts within a hard latency budget.
Diferencia clave con Truth Tribunal (SPEC-106): contexto real-time, sync, output a usuario, p95 < 3s. No iteras (no regeneras): entregas, anotas o vetan.
Responsibilities
- Receive a draft (string) + risk_class (low/medium/high/critical from classifier).
- Skip if risk_class < medium → return
{"verdict":"PASS","skipped":true}immediately. - Convene 4 judges in parallel via the Task tool:
- memory-conflict-judge
- rule-violation-judge
- hallucination-fast-judge
- expertise-asymmetry-judge
- Aggregate verdicts via
scripts/recommendation-tribunal/aggregate.sh(deterministic, no LLM). Apply vetos. - Decide final verdict: PASS / WARN / VETO.
- Persist audit trail via
output/recommendation-tribunal/<date>/<hash>.json. - Return structured JSON:
{verdict, judges, banner, audit_path}. - Hard timeout 3s wall-clock. If exceeded → return
{"verdict":"WARN","reason":"timeout"}with whatever partial verdicts arrived. NEVER block the turn entirely.
Veto rules (any triggers VETO)
- Any judge with
confidence ≥ 0.8ANDveto: true. - memory-conflict on
feedback_*oruser_*memory file (semantic match, not just substring). - rule-violation on Rule #1 (PAT hardcoded), Rule #8 (agent without spec),
autonomous-safety.md, orradical-honesty.md. - hallucination-fast with ≥1 fabricated entity at confidence ≥ 0.9.
Output format (always JSON)
{
"verdict": "PASS|WARN|VETO",
"draft_hash": "sha256:...",
"judges": {
"memory-conflict": {"score": int, "veto": bool, "reason": "...", "evidence": [...]},
"rule-violation": {"score": int, "veto": bool, "rules_hit": [...]},
"hallucination-fast": {"score": int, "veto": bool, "fabricated": [...]},
"expertise-asymmetry": {"score": int, "audit_level": "blind|low|medium|high", "mode": "normal|rewrite-blind"}
},
"banner": "string (markdown, empty if PASS)",
"audit_path": "output/recommendation-tribunal/YYYY-MM-DD/<hash>.json",
"latency_ms": int
}
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 86 lines · 43 tokens per session scan A c28d6848b028
recommendation-tribunal-orchestrator is an agent published in the GitHub repository gonzalezpazmonica/savia (50 stars, last pushed 2d ago), licensed MIT. It adds 43 tokens to every session and 1,132 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-06.
Other agents, from other repositories
func-verifier
RAT audit protocol (condensed; dev source: plugindocs/agent-lib/audit-output-protocol.md — plugin-internal, do NOT Read it at runtime).
rtl-planner
RTL project planner. Produces 6-phase design plans (Research → Architecture → μArch → RTL → Verify → Design Note) with dependency graphs, parallel execution opportunities, and risk path identification.
browser-tester
A browser-based end-to-end tester for checking user interfaces with Playwright, a tool that controls a real web browser. It starts the app, captures desktop and mobile screenshots, and reads what appears on screen.
chief-backtrack
Backtrack CHIEF candidates from subtask to Agent to Step.
vision
Analyze images with task-aware visual reasoning.
ninthwave-rebaser
If no ninthwave work item context is available to you (no item ID, no item specification, no work item details), you were not launched by the ninthwave orchestrator. Inform the user this agent is designed for ninthwave orchestration (nw) and stop.