aliatx2017/reasonix-hermes/research-report

Research Report — Summary Report

Trigger

/research-report

Workflow

Step 1: Locate Results

Find outline.yaml and fields.yaml in the current working directory or workspace root. Read the execution.output_dir from outline.yaml (default ./results).

Step 2: Read All Results

Use glob to find all .json files in the output directory. Read each JSON result file using read_file.

Step 3: Fill Gaps with Crawl4AI (Optional)

If any items have critical fields marked [not found] in their uncertain array, offer to deep-research them using crawl4ai-local:

curl -s -X POST http://192.168.1.214:11235/crawl \
  -H 'Content-Type: application/json' \
  -d '{"urls": ["<url>"], "priority": 10, "extract_markdown": true}'

Extract the missing field values and update the JSON with edit_file.

Step 4: Ask User About Summary Fields

Ask the user: "Which fields should appear in the table of contents summary line for each item?" Suggest fields that contain short, comparable values (e.g., dates, star counts, scores, ratings).

Step 5: Generate Report

Write report.md using write_file with this structure:

# Research: {topic}
Generated: {date}
 
## Table of Contents
1.[Item Name 1](#item-name-1) — {summary_field_1}: {value} | {summary_field_2}: {value}
 
## Item Name 1
### {Category 1}
-**{field_name}**: {value}
-**{field_name}**: {value}
 
### {Category 2}
-**{field_name}**: {value}
...

Report Generation Rules

1. Cover all fields — every field defined in fields.yaml must appear in the report
2. Skip uncertain values — skip any field whose value contains [uncertain] or whose name is in the item's uncertain array
3. Flag missing data — for [not found] fields, note "(data unavailable)" instead of omitting
4. Group by category — use the category grouping from fields.yaml
5. Format complex values:

• Lists of objects: format as sub-list or table
• Long text: use blockquotes for readability
• Nested objects: flatten with dot notation or sub-sections

1. Sources section — at the end of each item, include the _source URLs
2. Summary table — at the end of the report, add a comparison table with all items and a few key fields for side-by-side comparison

Step 6: Confirm

Show the report location to the user.

Step 7: Publish to Discord

Run the Discord publish script to post the report to the configured Discord channel:

bash .reasonix/scripts/discord-publish.sh ./report.md

The script automatically loads the webhook URL from .reasonix/.discord-webhook (gitignored). No environment variables are needed when the webhook file exists. The script splits the report into ≤1900-char chunks at paragraph boundaries and posts each as a Discord message.

If the webhook file is missing, set these environment variables:

• DISCORD_WEBHOOK_URL — a Discord webhook URL (preferred)
• DISCORD_BOT_TOKEN + DISCORD_CHANNEL_ID — bot token fallback

Output

• report.md — in the same directory as outline.yaml
• Discord messages — posted to the configured channel