work-resume

name	work-resume
description	Resume a task using fuzzy matching across all repos. Use when user wants to continue previous work, says "resume", "continue", or references a past task.
arguments	description

/work-resume

Agent Compatibility

This skill can run in Claude Code or Pi. Before running Workstation helper scripts, read ../_shared/agent-compatibility.md and resolve:

• WORKSTATION_DISPATCH → scripts/workstation-dispatch
• WORKSTATION_NOTIFY → scripts/workstation-notify

Use CLAUDE_PLUGIN_ROOT when it is set; otherwise resolve the script path relative to this skill directory.

Resume a task using fuzzy matching across all repos.

Implementation Steps

1. Scan all checkpoints

ls ~/.varie/sessions/*.yaml 2>/dev/null

If no sessions exist:

No saved sessions found.
Use /work-start <task_name> to begin tracking work.

2. Read and parse sessions

For each session file, extract:

• task.name and task.id
• task.tags
• current_step
• repo
• steps[*].notes (recent notes)

3. Fuzzy match against description

Match the user's description against:

• Task names (highest weight)
• Task tags
• Step names
• Step notes
• Repo names

Matching examples:

Input	Likely Match
"character api"	task.name contains "character" AND "api"
"frontend types"	step.name or notes contain "frontend" or "types"
"webapp"	repo = "webapp"

4. Handle match results

Single match: Proceed to load context.

Multiple matches: Present options:

Found multiple matches:

1. webapp: Character API Migration (step 3/5)
2. backend: Character Sync Feature (step 1/3)

Which one? (Enter number or type more specific description)

No match:

No matching task found for "{description}".

Active tasks:
- webapp: Character API Migration
- my-project: Communication Protocol

Try again with a more specific description, or use /work-status to see all tasks.

5. Load checkpoint and context

Read the matched session's checkpoint file.

Read the task's HANDOVER.md:

cat {repo_path}/{archive_path}/HANDOVER.md

6. Check if recovery needed

If last_active is old (> 1 hour), or git state doesn't match:

⚠️  Checkpoint may be stale (last: {time} ago)
Running recovery check...

Then run recovery logic (compare checkpoint vs actual git state).

7. Load into current session

Update the current session's context with:

• Task name and steps
• Current step and notes
• Files touched
• Git branch

8. Notify daemon

"$WORKSTATION_NOTIFY" session-start --task "{task_id}"

9. Output resume summary

Resuming: {Task Name} ({repo})
Session: {session_id}
Status: Step {N}/{total} in progress

Current step: {step_name}
Notes: "{last notes}"

Files touched:
- {file1}
- {file2}

Last checkpoint: {time} ago

Ready to continue. What would you like to work on?

Or if recovery was run:

Resuming: {Task Name} ({repo})

Recovery found more progress than checkpoint:
- {step X}: checkpoint said "in progress", actually complete
- Resuming from: {actual position}

Ready to continue from {description}.

Example

User: /work-resume character api

Output:

Resuming: Character API Migration (webapp)
Session: abc123
Status: Step 3/5 in progress

Current step: frontend_api_client
Notes: "Working on response type definitions"

Files touched:
- src/api/client.ts
- src/types/companion.ts

Last checkpoint: 15 minutes ago

Ready to continue. What would you like to work on?

Recovery Integration

If checkpoint is stale, automatically runs /work-recover logic:

Resuming: Character API Migration (webapp)

⚠️  Checkpoint from 2 hours ago. Checking actual state...

Recovery found:
  Checkpoint: "Working on response types"
  Git shows: types complete, fetch updated, hook pending

Adjusted: Continuing from useCompanions() hook (client.ts:145)