llm-observability skillA
llm-observability is agent-read markdown (skill) from nahid-sparktales/agent-dispatcher: See what an agent actually did — one trace per run with nested model, tool and retrieval spans, token and latency accounted per step, and failures clustered by mechanism instead of read one at a time. Use when an agent misbehaves in ways you cannot reproduce, when cost or latency is unexplained, when "it sometimes fails" is the whole bug report, or before writing evals when you do not yet know which failures exist. Not for judging whether an output is correct (agent-evals), and not a replacement.
Indexed from public GitHub and served as immutable, content-addressed versions. Install it pinned to an exact SHA-256 with the mdr CLI, and every file is verified against the hash recorded here before it reaches your agent. The deterministic audit below grades the latest version, and the same file always earns the same grade.
What the file says
# LLM observability An agent run is a tree of decisions, most of which nobody saw. Without a trace you are left re-running the prompt and hoping the failure happens again while you watch. ## When this fires A failure that will not reproduce, a cost or latency number nobody can account for, a "sometimes" bug, or the start of eval work where the failure modes are still unknown. It does not fire for a single deterministic call whose input and output you already hold. ## Procedure 1. **State the question first.** "Why did run X do Y", "where is the cost going", "which failures share a cause". Instrumentation added without a question becomes expensive noise nobody reads. 2. **One trace per run, nested by causality.** The run is the root; each model call, tool call, retrieval and retry is a child span in the place it actually happened. A flat log of lines loses the one thing you need — which step caused which. 3. **Record per span what cannot be reconstructed afterwards**: model id, prompt or system-prompt version, sampling parameters, input and output token counts, stop reason, tool name and …
Read the whole file at its exact version.
How to install
mdr add nahid-sparktales/agent-dispatcher/llm-observability@git:20260919.a0d4f55mdr add nahid-sparktales/agent-dispatcher/llm-observability@sha256:459396d48a7ea26aPin to a label to follow the author's releases, or to a sha256 to freeze the exact bytes forever. Either way the resolved hash is written to mdr.lock, and mdr install reproduces it on any machine.
[](https://markdownregistry.com/a/art_2r52wcctnzmdyddr)
1 badge views in 30 days
Versions
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (6276 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
nahid-sparktales/agent-dispatcher · 49 stars · license MIT · pushed 2026-09-23 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_2r52wcctnzmdyddr GET https://markdownregistry.com/api/v1/resolve?ref=nahid-sparktales/agent-dispatcher/llm-observability GET https://markdownregistry.com/api/v1/blob/459396d48a7ea26aa19ba96b9d5e97e23f6f1406e3413cd80672b92ad9ed4d8a
Your agent does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.