Output Workflow Trace File

---
name: output-workflow-trace-file
argument-hint: [workflow-name-or-run-id]
description: Read and render the output of a local Output SDK workflow trace file as clean readable markdown. Use when the user wants to view what a recent workflow produced, see the result from a local trace file, or render trace output as a document.
version: 0.1.0
---

Show just the final output of an Output.ai workflow trace — the actual result, rendered as readable markdown.

$ARGUMENTS - Either a workflow name (e.g. `context_competitors`) or a workflow run ID. If empty, use the most recent trace across all workflows.

## Instructions

1. **Find the trace JSON file:**
   - Trace files live in `logs/runs/<workflow_name>/` as JSON files
   - Filenames follow the pattern: `<timestamp>_<workflow_id>.json`
   - If a workflow name is given, find the latest `.json` file in `logs/runs/<workflow_name>/`
   - If a run ID is given, search across all `logs/runs/*/` folders for a file containing that ID in its filename
   - If no argument, find the most recently modified `.json` file across all `logs/runs/*/` folders

2. **Extract the output from the trace file.** You only need `output.output` from the JSON root — skip `children` and `input`.

   **Strategy for large files** (trace files can be 10k+ lines):
   - First, try `jq '.output.output' <file>` to extract directly — this is the fastest path
   - If `jq` is not available: read the **last 500 lines** of the file (the `output` field is at the root level, near the end of the JSON). Work backwards in chunks if needed
   - Do NOT read the entire file from the top — the `children` array with step details can be thousands of lines and you don't need any of it

3. **Save a markdown file to `tmp/trace_result_<workflow_name>_<id>.md`** (create the `tmp/` directory if it doesn't exist) with:

   ### Header (brief)
   - One line: workflow name, ID, duration

   ### Result
   Render `output.output` as clean, readable markdown:
   - String fields that contain markdown → render directly
   - Arrays of objects → render each as a sub-section with key fields
   - Arrays of strings → numbered lists
   - Nested objects → sub-sections with key-value pairs
   - URLs → render as links
   - Long text excerpts → render as blockquotes

   The goal is a document you'd want to READ, not debug. Make it look good.

4. **Tell the user** the saved file path.

## Important
- This is a RESULT view — render for readability, not debugging
- Do NOT include step details, inputs/outputs, or timing per step
- Do NOT wrap things in JSON code blocks — this should read like a document
- Include the full output without truncation