Research Agent Setup

Research Agent Setup

Interactive onboarding. Prompts the user for all configuration values and writes config.json.

When to Use

On first boot, or when the user asks to configure or reconfigure the research agent.

Process

Ask the user each question in order. Accept their answer before moving to the next. Write the completed config at the end.

Questions

1. Agent name -- What should this agent call itself? (default: Research Agent)
2. Timezone -- Your local timezone (e.g. America/New_York, Europe/London, Asia/Tokyo)
3. Daily brief time -- What local time should the daily research brief run? (default: 9am). The template's full schedule -- daily brief, evening topic briefing, weekly trends review, and weekly quality review -- is always installed; this only sets the daily brief time.
4. Target audience -- Who reads these briefs? (e.g. "technical founders building with AI", "content creators in the finance niche")
5. Niche / product -- What community, product, or topic is this agent serving? (used to tune business_fit scoring)
6. Source categories -- Which source types to enable? (GitHub, Reddit, YouTube, Hacker News, arXiv, RSS, custom URLs, Apify)
7. GitHub sources -- Topics to track (e.g. "ai-agents, llm-ops"), specific repos to monitor, or "none"
8. Reddit communities -- Subreddits to monitor (e.g. "LocalLLaMA, MachineLearning") or "none"
9. YouTube channels -- Channel handles or IDs to track via RSS or "none"
10. RSS feeds -- Any RSS/Atom feed URLs, comma-separated, or "none"
11. Hacker News -- Monitor HN top stories? (yes/no)
12. arXiv queries -- Search terms for arXiv (e.g. "agent memory, multi-agent systems") or "none"
13. Custom URLs -- Any specific URLs to fetch each run (changelogs, blog pages) or "none"
14. Apify actors -- Any Apify actor IDs to run (requires APIFY_TOKEN in .env) or "none"
15. Scoring rubric -- Use default rubric weights or customize? If customize: ask weight (1-10) for each dimension.
16. Top N signals per run -- How many signals to select and brief per run? (default: 8)
17. Min score threshold -- Minimum score to include a signal (default: 5.0)
18. Signal categories -- What categories to classify signals into? (e.g. "competitive intel, content opportunity, technical research, community signal")
19. Content voice -- Describe the content voice/style for brief writing (e.g. "direct, practitioner, no hype")
20. Delivery destination -- Where to send run summaries? (telegram / slack / local_markdown / none)
21. Delivery credentials -- If Telegram: use the agent's existing cortextOS Telegram channel. If Slack: webhook URL goes in .env, not config.
22. Require approval before sending externally? -- (yes/no, default yes)
23. Dedup window -- How many days back to deduplicate? (default: 7)
24. Retain raw dumps? -- Keep raw source dumps in research/output/? (yes/no, default yes)
25. Privacy exclusions -- Any topics, keywords, or sources to never include in output? (e.g. "competitor names, internal projects")

Output

Write or update:

• IDENTITY.md
• USER.md
• TOOLS.md
• research/sources.json
• research/scoring-rubric.json
• config.json

Crons -- install ALL of the template's shipped crons (do not reduce the set). The template ships 5 crons in config.json; preserve every one, each with its full skill-chain prompt and update-cron-fire call:

• heartbeat (4h)
• daily-research-brief (0 9 * * * -- adjust only the hour to the user's chosen brief time from Q3)
• topic-briefing (0 18 * * *)
• weekly-trends-review (0 9 * * 1)
• research-quality-review (0 12 * * 5)

Keep the shipped prompts verbatim. Only the daily-research-brief schedule changes to match the user's preference -- do NOT drop the other crons or replace them with terse run-only entries.

Write .env stub reminding user to add their API keys (if .env does not already exist). Use research.delivery.destination and research.delivery.requires_approval for delivery settings.

Never write secrets into committed files. Confirm: "Setup complete. Add your API keys to .env or org secrets. Your first run will fire at [next scheduled time]."