knowledgeintermediate

Documentation generator

A specialised coding agent pointed at your source tree. Reads modules, infers behaviour, updates docs/ to match. Writes through the workspace mirror so changes hit your repo as a clean PR you can review before merging.

lines 36cost ~$0.30 per pass on a small repoinstall hub://digitorn/doc-writer

Install in one command

digitorn install hub://digitorn/doc-writer

What you get

Doc PRs against your repo, ready to merge
Code blocks in docs that match the source they describe
Read-before-edit guard, no hallucinated function signatures
Atomic per-file PRs, easy to review

Credentials this template uses

Anthropic Claude

Click each one for the credential setup command, scope guidance, and provider details.

The full YAML, ready to fork

app.yaml

1app:2  app_id: doc-writer3  name: "Doc writer"4  version: "1.0.0"5  category: "knowledge"67execution:8  mode: conversation9  entry_agent: writer1011modules:12  filesystem:13    config:14      read_before_edit: true15  workspace:16    config:17      render_mode: markdown18      entry_file: docs/index.md19      sync_to_disk: true20      lint: true2122agents:23  - id: writer24    modules: [filesystem, workspace]25    brain:26      provider: anthropic27      model: claude-sonnet-4-628      credential:29        ref: anthropic_main30        scope: per_user31        provider: anthropic32    system_prompt: |33      Read modules under src/. Update docs/ to match.34      One file at a time. Read before you write.35      Code blocks in docs must match the source exactly.

How it works

workspace render_mode markdown

The workspace serves a markdown preview while the agent works. sync_to_disk: true mirrors writes to your repo so the result is a real file change you can git diff.

filesystem + workspace together

filesystem for reading the source, workspace for writing the docs. Same _read_files set is shared, so the read-before-edit guard prevents the agent from writing about a file it has not opened.

system prompt enforces atomic edits

'One file at a time. Read before you write.' The agent reads a source module, updates the matching doc page, moves on. Atomic per-file edits land as separate commits in the resulting PR.

Read the deep dive

10 apps you can ship in 50 lines of YAML

Read article

Newsletter

Get the next post in your inbox.

Engineering notes from the Digitorn team. No marketing, no launch announcements, no "10 prompts that will change your life". Just the things we write that we'd want to read.

Other templates

developerStarter coder/templates/starter-coder developerClaude Code clone/templates/claude-code-clone researchDeep research agent/templates/deep-research messagingSlack helper/templates/slack-helper developerPull request reviewer/templates/pr-reviewer opsWeekly digest reporter/templates/weekly-digest creativeLive React app builder/templates/react-builder messagingVoice assistant/templates/voice-assistant opsScheduled data pipeline/templates/data-pipeline