momentic-result-classification — Claude Code skill: triage failed
Reads failed Momentic runs through MCP, pinpoints the step where the run first went wrong, explains why, and files the verdict under a category.
npx skills add momentic-ai/skills --skill momentic-result-classification -a claude-codeWhat momentic-result-classification does
This skill wires the agent to Momentic's MCP tools. It pulls run metadata, a single step result with before and after screenshots, the trace, plus a snapshot of the test steps. From there it hunts for the moment the page state first diverged, instead of settling on the last assertion that blew up or the locator that went missing.
The investigation order is spelled out: current run first, then repeated module invocations inside that same run, and only after that older runs for comparison. Screenshots are treated as the primary record of page state; trace and DOM exist to explain why the picture changed. A container that passed while a child step failed counts as a partial failure until its closing screenshot gets checked. Vague causes like an unreliable setup do not pass — the skill wants the broken postcondition named outright.
Momentic itself is an end-to-end framework: tests are YAML steps that mix natural-language checks with Playwright actions. Run data lands in a test-results folder holding metadata.json and a zip per run. You need the Momentic MCP server reachable. The final verdict goes back through momentic_submit_result_classification, so the category ends up in the run data rather than staying in the chat.
What it helps with
- locate the earliest step where a run left a good state
- explain a failure at action level, not at the final assertion
- bucket a failure into a category and persist the verdict
- compare the current run against earlier runs on the same git branch
- tell a real container failure from a recovery that masked it
Example prompts for the agent
- Work out where this run first broke and explain the root cause
- Classify this failure and submit the result to Momentic
- Compare this run with previous ones on the same branch — is the test flaky?
How to install momentic-result-classification
1Skills CLI (Vercel)
npx skills add momentic-ai/skills --skill momentic-result-classification -a claude-code2Manually: put the files in place
~/.claude/skills/momentic-result-classification/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/momentic-result-classification/
1Skills CLI (Vercel)
npx skills add momentic-ai/skills --skill momentic-result-classification -a codex2Manually: put the files in place
.agents/skills/momentic-result-classification/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add momentic-ai/skills --skill momentic-result-classification -a cursor2Manually: put the files in place
.cursor/skills/momentic-result-classification/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install momentic-ai/skills momentic-result-classification2Manually: put the files in place
.github/skills/momentic-result-classification/SKILL.md1Skills CLI (Vercel)
npx skills add momentic-ai/skills --skill momentic-result-classification -a gemini-cli2Platform CLI
gemini skills install https://github.com/momentic-ai/skills --path skills/momentic-result-classificationManually: put the files in place
.devin/skills/momentic-result-classification/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add momentic-ai/skills --skill momentic-result-classification -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about momentic-result-classification
How do I install momentic-result-classification in Claude Code?
Run npx skills add momentic-ai/skills --skill momentic-result-classification -a claude-code. The skill file then sits at ~/.claude/skills/momentic-result-classification/SKILL.md.
What does momentic-result-classification do?
It guides the agent through a failed Momentic run: find the first divergent step, back the cause with screenshots, trace and DOM, then assign one category and record the verdict through MCP.
What does the skill need to work?
A reachable Momentic MCP server exposing tools such as momentic_get_run and momentic_get_step_result. Without it the skill has no run data to inspect.
