work-update

name	work-update
description	Sync work context from external sources, update tasks and memory, and report changes

Work Update

Refresh tasks and memory from ongoing work activity.

Load memory-management and task-management before continuing. This command assumes /work-start has already initialized TASKS.md, CLAUDE.md, and the memory wiki.

Flow

1. Load State

Read:

• TASKS.md
• CLAUDE.md
• memory/index.md
• memory/glossary.md

Read when needed:

• memory/log.md
• specific pages under memory/

If TASKS.md, CLAUDE.md, or core wiki files are missing, suggest /work-start.

2. Sync Tasks From External Sources

Before checking a source, look for and load the relevant skill from .claude/skills when available.

Check available task sources such as:

• project trackers like Asana, Linear, or Jira
• GitHub Issues: gh issue list --assignee=@me

Compare external tasks against TASKS.md:

• new external task -> offer to add
• matching task already tracked -> skip
• task missing externally -> flag as potentially stale
• completed externally -> offer to mark done

TASKS.md Confirmation Rules

Do not edit TASKS.md directly from scanned activity unless the user confirms the change.

During task sync and activity scan:

• Collect candidate task additions, completions, stale items, and waiting-state changes.
• Compare each candidate against existing TASKS.md entries.
• Present a short triage prompt before editing.
• If the user chooses no changes, leave TASKS.md untouched.
• Only apply confirmed task edits.

Safe without confirmation:

• Reading TASKS.md
• Reporting possible stale tasks
• Reporting likely missing tasks
• Reporting external completion signals

Requires confirmation:

• Adding a task
• Completing a task
• Moving a task between Active, Waiting On, Someday, and Done
• Changing task wording, due dates, links, or subtasks

3. Scan Recent Activity

Before checking a source, look for and load the relevant skill from .claude/skills when available.

Optional Project Subagents

If the Agent tool is available and project-local agents exist under .pi/agents/, prefer using them for read-only collection and classification when the scan spans multiple sources or would otherwise consume substantial context.

Project-local agents useful for this workflow:

• activity-scanner - general read-only scanner for Slack, Gmail, Calendar, Drive, Notion, or a specific project/source scope
• task-reconciler - compares candidate tasks against TASKS.md and prepares task triage recommendations
• memory-curator - classifies findings for durable memory updates

Use subagents this way:

1. Before spawning new scanners, check the current conversation for completed activity-scanner outputs, especially scheduled hourly runs.
2. Reuse existing scanner outputs when their ## Scanner metadata shows a useful run time and relevant sources scanned.
3. Spawn new activity-scanner agents only for missing sources, stale/low-confidence outputs, or explicit user refresh requests.
4. Aggregate reused and newly scanned outputs in the parent agent.
5. Pass the aggregated task-like findings to task-reconciler before asking the user to confirm task edits.
6. Pass the aggregated memory-like findings to memory-curator before applying safe memory updates or asking for confirmation.

Reusing Hourly Scheduled Scanner Results

If activity-scanner runs hourly and the user runs work-update at end of day, do not treat older same-day scanner outputs as stale merely because they are more than a few hours old. Instead:

• Reuse all visible completed scanner outputs from the current workday, or since the last known work-update, when their metadata shows relevant sources scanned.
• Deduplicate repeated findings across hourly runs.
• Identify gaps by source.
• Run a small catch-up scan only for sources not covered by scheduled scans or when recent scanner outputs are missing/stale.
• If scanner outputs are not visible in the current conversation and no agent IDs are available to retrieve, fall back to normal scanning.

When reusing scanner outputs, mention it in the final report, for example:

Reused scanner outputs:
- Slack/email/calendar scanner runs from 09:00-17:00

New scans:
- Catch-up scan because recent scanner outputs were missing/stale
- Recent docs, because scheduled scans did not cover Drive/Notion

Subagents must only collect and classify. The parent agent remains responsible for user confirmation, edits to TASKS.md, edits to CLAUDE.md or memory/, and running memory-backup when required.

Gather recent signals from available sources:

• chat
• email
• documents
• calendar
• meeting notes only when provided by the user or available through a supported readable/exported source

Use the activity scan to surface:

• task updates
• likely missing tasks
• durable memory candidates
• status, ownership, relationship, or terminology changes worth preserving

4. Flag Missing Or Stale Tasks

Surface likely todos not tracked in TASKS.md.

Review active tasks for:

• past due dates
• 30+ day age
• missing project or person context
• external completion signals

Present triage options such as:

• done
• reschedule
• move to someday
• keep active

These triage options are suggestions until the user confirms the TASKS.md edit.

5. Resolve Unknowns

For each task or activity item, use the memory lookup order:

1. CLAUDE.md
2. memory/index.md
3. memory/glossary.md
4. relevant pages under memory/
5. memory/log.md if recent changes matter

Track what is known and what still needs clarification.

Ask about unknown terms, people, projects, or context only when the lookup order does not resolve them.

6. Ingest Durable Updates

Follow memory/SCHEMA.md when creating or updating durable wiki pages.

Process updates in this order:

1. update existing pages when the new information fits naturally
2. add meaningful markdown links when creating new cross-references
3. create a new page only when the information does not fit an existing page cleanly
4. file substantial reusable outputs into memory/queries/ or memory/comparisons/ when appropriate
5. update memory/index.md for every new, renamed, deleted, or reclassified durable page
6. append the action to memory/log.md

Useful enrichments include:

• links added to existing project, person, topic, or context pages
• clear status changes
• durable relationships or ownership notes
• recurring terms added to memory/glossary.md
• deadlines or milestones worth preserving in project context
• reusable comparisons or synthesized answers

Keep memory compact:

• Merge or replace stale bullets instead of appending routine daily updates.
• Do not duplicate standup details across person and project pages.
• People pages should keep stable role, current work areas, and durable milestones only.
• Skip transient logistics, repeated progress, and implementation detail with no future decision value.

For recurring query or comparison outputs, prefer a stable filename and update the existing page instead of creating a dated duplicate.

7. Apply Confirmation Rules

Safe to apply during work-update:

• enrich an existing memory page
• add a clear glossary entry
• add links or relationships to an existing page
• update memory/index.md or memory/log.md
• file a clearly valuable durable update into an existing page

Still require confirmation:

• new person page
• new project page
• new context page
• new topic page
• new comparison or query page when the value is uncertain
• weakly supported term promoted into CLAUDE.md

If durability is unclear, present a suggestion instead of editing memory.

8. Report

Summarize:

• task changes
• hot-cache changes
• wiki changes
• filed outputs
• unresolved gaps

Example structure:

Work update complete:
- Task changes: X added, X updated, X stale
- Hot cache changes: X
- Wiki changes: X pages updated, X pages created
- Filed outputs: X queries, X comparisons
- Unresolved gaps: X

If memory changed, say whether it was:

• a safe update to an existing page
• a newly created page after confirmation
• a filed query or comparison
• a suggested but unconfirmed addition
• a gap that still needs clarification

9. Memory Backup

If this run changed CLAUDE.md, TASKS.md, or anything under memory/, trigger memory-backup.

Notes

• This command is interactive.
• External links should be preserved when available.
• Fuzzy title matching is fine for task comparison.
• Prefer updating existing pages over creating new ones.
• Use /work-start to initialize missing task or memory foundations.
• If nothing changed, skip backup.