Agent skill

Everything on this site, packaged so your coding agent can act on it. One command:

npx skills add https://github.com/Jwuthri/Tracely --skill tracely

Installed via the open-source skills CLI — Claude Code, Cursor, Copilot, Antigravity and anything else it targets. -g installs it for every project on the machine; --agent '*' installs it into every agent you have. It is plain Markdown with YAML frontmatter, so copying skills/tracely/ into your project by hand works too.

What it carries

A short entry point plus five references the agent loads only when the task calls for them:

Reference	Covers
`automatic.md`	`instrument="auto"`, provider + framework extras, `trace()`, `@observe`, drop-ins, LangGraph, LiteLLM, agent SDKs, redaction, threads, and the OTLP conventions for non-Python services.
`manual.md`	The full manual span API — every observation type, handoffs, RAG, state deltas, multimodal I/O, and the record-replay seam.
`evaluators.md`	Evaluation columns: structural vs judge, levels, `@VARIABLE` templates, advisory verdicts, sequential grading, targeting and sampling.
`ci-gate.md`	Scenarios, adversarial suites, hermetic replay and the CLI + GitHub Action.
`troubleshooting.md`	Symptom → cause → fix, ordered by how often it’s the answer.

The entry point leads with the handful of conventions that fail silently when you get them wrong — a missing conversation id, a swallowed tool error, no flush(), a dropped traceparent, the inverted polarity of an adversarial scenario. Each one produces a green-looking workspace that can’t answer the question it was set up for.

Skill vs MCP server. The skill gives your agent the know-how; the MCP server gives it your data. They compose — running both is the intended setup, and it’s what turns “add tracing here” into “look at the last 20 traces, work out what’s failing, and add a column that catches it.”