A read-only research agent that maps the codebase
A read-only research agent that maps the codebase, answers architecture questions in chat with cited file paths, and ships a weekly written explainer that grows into durable team knowledge.
Run9 steps1-2 days build, ongoing weekly use
Run a research agent on your codebase
A read-only research agent that maps the codebase, answers architecture questions in chat with cited file paths, and ships a weekly written explainer that grows into durable team knowledge.
Steps
Pointers
Codebase research plan
Question log
+1 moreTitleStatusOwner
Pick the codebase + define the read-only scopeshippedYou
Seed the architecture map with a top-down passactiveAgent
Set up the question pipeline + logqueuedYou
Define the question taxonomy: what the agent will + won't answerdraftedAgent
…
Public preview of the cloned workspace. Same chrome you'll see after Open in Dock.
Spin up an agent for the heavy lifting
Reads the codebase + answers questions, never edits files.
9 steps, 11 official links, 4 agent prompts
Every external doc the agent needs to cite is pre-loaded into the workspace's Pointers table. No hunting for the right URL mid-draft.
What's inside
Pre-loaded so day one is execution.
5Surfaces
9Steps
4Agent prompts
11Official links
5Tools mapped
Surfaces
- tableSteps
- tablePointers
- docCodebase research plan
- tableQuestion log
- docStatus
How the loop works
Your agent works. Dock shows you what happened.
Open this template and you get a workspace seeded with an agent prompt. Connect your agent — Claude via our MCP, Cursor, your own setup — and it reads, drafts, and posts updates as it goes. You watch Dock for the latest.
- 01
Connect your agent
Claim an agent invite at trydock.ai/agent-invites — your agent gets an API key scoped to this workspace. Paste the key into Claude Desktop, Cursor, or any MCP client.
- 02
Your agent reads the workspace
The agent prompt at the top of the workspace tells your agent its role, the cadence to follow, and the surfaces to update. No extra setup — open Dock and your agent already knows what to do.
- 03
Watch Dock for the latest
Your agent posts to the Status surface after every meaningful action — newest at top. Wire the workspace's webhooks to Slack or email to get pinged in real time.
Wire it up · Claude Desktop
Add Dock as an MCP server in 30 seconds.
{
"mcpServers": {
"dock": {
"command": "npx",
"args": ["-y", "@trydock/mcp"],
"env": {
"DOCK_API_KEY": "<paste from /agent-invites>"
}
}
}
}Drop into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on Windows / Linux. Restart Claude Desktop. Ask Claude:“Read trydock.ai/<org>/run-a-research-agent-on-your-codebase and follow the agent prompt.”
FAQ
Common questions on this template.
Why a separate research agent vs. just using Claude Code or Cursor for everything?
Scope discipline. A multipurpose 'do anything in the repo' agent will edit a file you didn't ask it to edit, eventually. A research agent in read-only mode never can. The boundary is the safety. You can always spawn a separate coding agent (with write access) when you've identified the change to make; the research agent's job is to find it without changing it.
How do I keep the agent from hallucinating module relationships?
Three rails: (1) demand a citation (file path + line number) for every claim, (2) calibrate confidence buckets in the system prompt and spot-check the high-confidence claims weekly, (3) run a top-down architecture pass first, refer back to it instead of letting each question be a fresh exploration. The combination keeps hallucination rates under 5% on high-confidence answers in practice.
Should the agent commit its own explainers to the docs repo?
No. Edit + commit is a human step. The agent drafts in the Brief surface; you review, edit, and commit. Auto-committing AI-drafted docs is how teams end up with 200 docs nobody trusts. Manual commit is the trust gate.
What runtime should I pick: Claude Code, Cursor, or a custom MCP-based agent?
Claude Code is the simplest default: terminal-native, full filesystem access, supports a --no-write flag for read-only mode. Cursor is the right choice if you're already in the editor and want in-context Q&A. A custom MCP-based agent (with the workspace + Slack + git MCPs wired in) is more work but gives you a permanent agent that lives in chat, not a per-session conversation.
What does this cost in API tokens?
A typical research session (5-10 questions, 1 architecture pass) is ~$0.10-0.50 on Claude Sonnet. A weekly explainer is ~$0.20-1. Total monthly cost for a team of 5-10 asking the agent 20-50 questions/week: $5-30/month. Heavy users with the agent in a Slack channel: $20-80/month. Negligible compared to the time saved.
Can my AI agents help build the agent?
Yes. The template ships agent prompts for the slow parts: the top-down architecture map, the confidence calibration pass, the weekly explainer drafts, and the Monday-review process. The Question log surface is the canonical record, every question logged with citations + confidence, every weekly explainer linked back to the questions it summarized.
Related templates
More from Run
RunMixed
Launch on Product Hunt the right way
12-step plan from picking the launch date to shipping the post-mortem. Your agents draft the assets + run the 24-hour hunt log + flag every converting comment in real time.
RunMixed
Launch on Hacker News (Show HN guide)
10-step plan for landing on the HN front page with a Show HN. Real title rules, real comment-thread tactics, real survival guide.
RunMixed
Automate Instagram posting with your agents
10-step plan for a posting pipeline that drafts, queues, and ships 6-12 Instagram posts a week with agents drafting and a human approving.
Used by these teams · 2
Teams that run this template, with their full setup.
Data analyticsUse case
Ad-hoc analytics asks stop drowning in Slack.
Requests, analyses, dashboards in one workspace. The SQL + reasoning + result lives next to the request, so the next similar question gets a 2-min copy-edit instead of a re-derivation.
Open the use case
Competitive intelligenceUse case
Beat the reactive 'we should do that' cycle.
Competitor updates auto-collected nightly. A weekly implications doc. Strategy emerges instead of being chased. The indexer reads competitor blogs/changelogs/X; the writer drafts the 'should we?' essay weekly.
Open the use case
Open it. Hand it to your agent. Ship.
One click mints a fresh workspace in your org with the template body seeded. Your agents, your team, your edits from there.
About this template
Curated by the Dock team at Vector Apps Inc. Every template is a real shared workspace we run with our own agents before publishing.
Reviewed regularly by the Dock team. Each playbook step links to the upstream tool's official docs so we can re-verify the rules as platforms change.