Kelly — Documentation Keeper

The problem#

Projects evolve fast. Maxime + the crew ship features, refactor, change conventions. Docs rot in days. New agents (or future-Robbie post-context-loss) burn cycles re-discovering things that were already figured out. Docs are the single biggest leverage point we're under-using.

The agent#

🌺 Kelly — Documentation Keeper. Archivist and curator of all crew knowledge. Obsessive about saving everything the team produces, labelling it, and making it findable. Reads code and writes it back into prose. Reviews every recent commit. Saves every deliverable, report, and decision into the permanent archive.

• Id: kelly
• Workspace: ~/.openclaw/workspaces/kelly
• Model: deepseek/deepseek-chat (DeepSeek V4-Flash — selected for large-context code reading accuracy)
• Color: dusty rose (#A0344A)
• Emoji: 🌺

What Kelly does#

1. Watches projects — for each project under ~/projects/<repo>/ and ~/.openclaw/workspace/projects/<slug>/, tracks the last commit she's seen.
2. Reviews diffs — on schedule, compares current state vs last-seen, identifies feature/API/contract changes that affect docs.
3. Updates docs in place:
   • README.md — features list, install, quick-start
   • docs/architecture.md — high-level system map, data flow
   • docs/api.md — API endpoints, server actions, request/response shapes
   • docs/conventions.md — file layouts, naming patterns, gotchas
4. Cross-project knowledge index — maintains ~/.openclaw/workspace/docs/INDEX.md listing every project + one-line summary + link to its docs. Other agents read this first.
5. Archive & curate — saves every artifact the team produces to projects/_archive/. Maintains the archive INDEX.md, labels everything clearly, purges duplicates and noise.
6. Reports — posts a daily summary to her Discord channel: what changed, what she updated, what was archived.

Trigger#

• Daily 03:00 SGT scheduled cron — review yesterday's commits across all watched repos, update docs, commit.
• Post-merge hook (later) — runs immediately after any PR merges to main on watched repos.

Channels#

• New Discord channel #kelly under OpenClaw-private guild — needs:
  • Channel created (Maxime in the Discord UI)
  • Channel ID added to ~/.openclaw/openclaw.json bindings[] mapping to agentId: "kelly"
• Same WhatsApp relay protocol as the others (no direct number, posts via Robbie's relay tag [notify|kelly] ...)

What's in scope (initially)#

• ~/projects/mission-control — Next.js app, biggest moving target
• ~/.openclaw/workspace itself — meta-docs, conventions, RELAY-PROTOCOL
• ~/.openclaw/workspaces/{debbie,chris,kelly} — each worker's local notes
• (Future) any new project that gets a PROJECT.md

What's out of scope#

• Doesn't write code. Reads it, documents it. If something needs a code change to make sense, she opens a task and assigns it to Chris.
• Doesn't make stuff up. If she can't tell what something does from the code/commits, she flags it and asks.

Tasks#

• t-046: Create Kelly's agent (config, workspace dir, identity, soul, team file)
• t-047: Add Discord channel binding for Kelly (after Maxime creates the channel)
• t-048: Build Kelly's daily review skill — diff watcher + doc updater
• t-049: Schedule the 03:00 daily cron job
• t-050: Bootstrap initial docs for mission-control, openclaw-workspace, local-llm projects (Kelly's first runs)

Open questions for Maxime#

• Burgundy / dusty rose for color — confirm or pick another?
• Should Kelly auto-commit doc changes, or open PRs and wait for review? (default: auto-commit on docs/ paths only, never touches code)
• WhatsApp summaries every day, or only when "interesting" changes detected?