Intern Wiki Agent

A Slack-native knowledge-management agent for tracking intern activity, built on LangGraph. It ingests updates posted in Slack, maintains a structured local markdown wiki, answers questions about who did what, and posts scheduled follow-up reminders.

Architecture

Three independent skills, connected only through a shared local markdown wiki (no tight coupling):

1. Writing -- parses an update, resolves the person, merges it into their page, self-checks for fabrication, and commits atomically. Also auto-updates project rosters, the index, and a follow-up list.
2. Querying -- answers questions using a hierarchical fallback search: index.md -> person page -> wikilinks -> raw minutes -> raw transcript.
3. Follow-up -- a scheduled (cron / Task Scheduler) job that reads the follow-up list, categorizes items by due date, and posts a report to Slack.

Design principle

Anything correctness-critical (rosters, wikilinks, index sync, frontmatter, identity resolution) is deterministic Python. The LLM only writes prose. This avoids run-to-run LLM unreliability affecting the data layer.

Question types (Querying)

• list_people -- who is on a project (reads the project roster)
• collaborators -- who a person shares projects with
• activity_on_date -- what someone did on a specific date (reads dated History)
• activity_general -- what someone has been doing (reads Summary)
• open_items -- what is pending for someone (reads the follow-up list)
• status -- current status of a person or project

Stack

• LangGraph -- StateGraph per skill
• Slack Bolt (Socket Mode) -- Slack integration
• OpenRouter -- LLM access
• thefuzz -- fuzzy name/project resolution
• Local markdown files -- storage (no external DB)

Wiki structure

See wiki-sample/ for the layout. Each person and project is a markdown file with YAML frontmatter, a Summary, and a dated History. Projects also have a deterministic People roster.

Setup

1. python -m venv .venv then .venv\Scripts\activate
2. pip install -r requirements.txt
3. Copy .env.example to .env and fill in your Slack + OpenRouter credentials.
4. python main.py to start the live listener.
5. Schedule python -m src.followup_agent (cron / Task Scheduler) for daily follow-up reports.

Note

The live wiki/ folder (real intern data) is gitignored. A wiki-sample/ is included to demonstrate structure.