Skip to content

Context Collection Directions

How to collect context and synthesize an orientation — priority order, flags, contract cross-reference

When to use

user wants to know their current work state or collect context

Slash command: /wb-context-collect

Linked workflow: context/collect-and-orient

  • context/context_bundle
  • context/context_git
  • context/context_tasks
  • contracts/active_contracts

Directions

Use mcp__work-buddy__wb_run("context_bundle") to collect. Scope options: - Default: config.yaml windows - Quick: {"hours": 24} - Last N days: {"days": 3} - Individual: context_git, context_chat, etc.

Synthesis instructions

Read each context file in priority order (git -> tasks -> projects -> obsidian -> chats). Extract the signal -- don't dump raw contents.

Present a concise orientation covering:

Active work (from git): - Which repos had recent commits? What kind of work? - Any dirty working trees? (uncommitted changes = in-progress work)

Projects (from projects summary): - What projects are currently active? State? - Do active projects align with where time is spent?

Current state (from Obsidian): - Journal entries today? Sign-in data? - Incomplete tasks? Any tagged as focused?

Recent conversations (from chats): - Last few sessions about? Any unfinished work?

Flags -- surface when data warrants: - No journal entry today -> user hasn't checked in - Running Notes growing without triage -> insight hoarding - Git activity in repos not related to any contract -> potential drift - Lots of infra commits, no paper artifacts -> infrastructure displacement - Projects with no recent activity -> stale or abandoned?

Contract cross-reference

Cross-reference against active_contracts: - Does git activity align with any active contract? - Is there work happening that doesn't map to any contract?

If no contracts exist: 'No active contracts -- all work is currently unanchored.'

Suggest one next action

Based on orientation, suggest one concrete next step -- highest-leverage action given current state.

Output format

Keep to 10-15 lines max. The user doesn't need a report -- they need a mirror.

Don'ts

  • Don't read every file verbatim
  • Don't generate a 50-line report
  • Don't invent concerns not in the data
  • Don't skip the contract cross-reference
  • Don't import work_buddy.* modules when a wb_run capability exists