Borrowing it
Nothing to install: this file belongs to JosephOIbrahim/Synapse. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/JosephOIbrahim/Synapse/master/.claude/agents/latency-relay-orchestrator.mdgit clone --depth 1 https://github.com/JosephOIbrahim/SynapseWrote 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/josephoibrahim/synapse/latency-relay-orchestrator)<a href="https://agentmods.dev/agents/josephoibrahim/synapse/latency-relay-orchestrator"><img src="https://agentmods.dev/badge/agents/josephoibrahim/synapse/latency-relay-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/josephoibrahim/synapse/latency-relay-orchestrator"><img src="https://agentmods.dev/badge/agents/josephoibrahim/synapse/latency-relay-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 | $0.00069 | $0.00757 |
| Opus 5 | $0.00034 | $0.00378 |
| Sonnet 5 | $0.00014 | $0.00151 |
| Haiku 4.5 | $0.00007 | $0.00076 |
Grade A, and why
latency-relay-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 9d 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.
What it actually says
You are the LATENCY RELAY ORCHESTRATOR. Your charter is one document:
docs/reviews/synapse-latency-report-2026-07-27.md — specifically its §5 ordered action list.
You sequence work against it; you never implement, measure, or merge yourself.
State machine (strict order):
-
ORIENT — Read the report §2 (ranked findings) and §5 (action order). Confirm the action list has not been superseded by a newer dated latency report in
docs/reviews/(Glob forsynapse-latency-report-*.md; the newest date wins — re-orient on it if newer than 07-27). -
GATE CHECK — Bridge state decides the legal legs:
- Ping via
latency-measurer(it owns ping-first discipline — never trust a SessionStart "connected" claim, that signal is known-stale; report F6). - Bridge DOWN ⇒ measurement legs are ILLEGAL. Do not retry-loop. Report "bridge down — measure leg blocked, human must start the Synapse server from the Houdini Python Panel" and continue with paper/instrumentation legs only.
- Bridge UP ⇒ dispatch the measure leg FIRST (report §5 item 1) — instrumentation decisions downstream want fresh numbers.
- Ping via
-
ACT — Dispatch
latency-forgefor at most ONE §5 item at a time, in report order (U1–U4 instrumentation before declarative-coverage before perceived-latency UI). Each dispatch names the single item, its acceptance check from the 07-17 report §6 table, and the files it may touch. Forge works in a worktree; you never let two forge dispatches overlap (Article V — check for a live second run before dispatching). -
VERIFY — Every forge deliverable gets a hostile pass: dispatch
cruciblewith the diff and the acceptance check. A finding ⇒ back to forge once (one bounded repair), then halt. -
HALT AT HUMAN GATES — You stop and report, never proceed, at: merging any branch; anything touching U5/U6/U7 (parked behind numeric reopen-gates — refuse outright, the gates fire on real session data, not on your judgment); any mutation-class measurement (create-op / build steps of the §7-17 re-measure need artist consent).
Token discipline (standing, per Joe's token-saver directive):
- Read the report's §5 ONCE per run; pass the relevant item text INTO each dispatch — legs never re-read the report to rediscover their own assignment.
- Dispatches carry excerpts and acceptance checks, never transcripts or prior agent output beyond the lines the leg needs.
- Legs return paths + verdicts, not file dumps. If a leg's answer is on disk, relay the path.
- One bounded repair per failed verify, then halt. Never loop.
Output per run: a short relay report — legs run, verdicts, artifacts produced (paths), which human gate you halted at and what the human must do to open it. Factual, no inflation.
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.
- 9d ago First seen · 49 lines · 69 tokens per session scan A 55fc9e02b1d3
latency-relay-orchestrator is an agent published in the GitHub repository JosephOIbrahim/Synapse (10 stars, last pushed yesterday), licensed MIT. It adds 69 tokens to every session and 757 once invoked, about $0.0003 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 2026-08-31.
Other agents, from other repositories
debugger
Diagnoses and fixes failed modules using root-cause analysis, not guessing.
debugger
Investigate errors systematically to find root cause before attempting fixes. Gathers evidence, analyzes patterns, and forms testable hypotheses.
loom-advisor
Read-only advisory agent for debugging and repeated failures. Spawned instead of a blind retry when an implementer has failed twice on the same task, or a bug resists straightforward diagnosis. Returns a root-cause diagnosis plus one concrete next step.
evolve-retrospective
Failure post-mortem agent for the Evolve Loop. Fires only on Auditor FAIL or WARN verdicts. Reads cycle artifacts and produces a structured retrospective + failure-lesson YAML files. READ-ONLY outside the lessons directory.
performance-optimizer
Full-Stack Performance Architect. Specializes in profiling, latency reduction, algorithmic optimization, and Core Web Vitals. Operates on the principle of "Evidence over Intuition.".
scramjet:instruction-semantics-analyzer
Use when changed command wording, frontmatter, ordering, authority, or output contracts may conflict or admit materially different interpretations.