resume-session

maslennikov-ig's avatarfrom maslennikov-ig

Resume a previously saved session by loading context and recent log entries. Use at start of orchestrator workflows to check for existing sessions.

⭐1stars🔀0forks📁View on GitHub🕐Updated Jan 9, 2026⬇️1downloads
[Memory and Context Management]

When & Why to Use This Skill

The resume-session skill is a vital utility for Claude agentic workflows, designed to maintain continuity by restoring previously saved session states. By loading historical context, recent logs, and workflow phases from local storage, it enables agents to pick up exactly where they left off, preventing redundant work and ensuring a seamless user experience across long-running tasks.

Use Cases

  • Resuming complex multi-stage coding or debugging workflows after a break or system restart to maintain progress.
  • Recovering agent state and decision history following an unexpected connection timeout or context switch.
  • Efficiently switching between different project environments by reloading specific workflow contexts and git states.
  • Automating the 'Phase 0' check in orchestrator workflows to determine whether to start fresh or continue an existing session based on data freshness.
nameresume-session
descriptionResume a previously saved session by loading context and recent log entries. Use at start of orchestrator workflows to check for existing sessions.
triggerAt the start of health workflows, or when user asks to continue previous work

Resume Session

Checks for and loads a previously saved session context, enabling seamless workflow continuation.

When to Use

  • At the start of health orchestrator workflows (automatic)
  • When user explicitly asks to continue previous work
  • After session restart or context switch

When NOT to Use

  • For new workflows with no prior context
  • When user explicitly wants to start fresh

Algorithm

Step 1: Check for Existing Session

Check if .tmp/current/session/context.md exists.

Step 2: Validate Freshness

If file exists, check the "Updated" timestamp:

  • If < 24 hours old: Session is valid
  • If > 24 hours old: Session is stale, ask user

Step 3: Load Context

Read .tmp/current/session/context.md and extract:

  • Current workflow and phase
  • Last action and next steps
  • Git state
  • Quality gate status

Step 4: Load Recent Log Entries

Read last 10 entries from .tmp/current/session/log.md:

  • Recent decisions
  • Issues encountered
  • Learnings

Step 5: Present Resume Option

If valid session found, present to user:

Found previous session:
- Workflow: health-bugs
- Phase: 3/7 (Staged Fixing)
- Last action: Fixed auth validation
- Age: 2 hours ago

Options:
1. Resume from where you left off
2. Start fresh (archive current session)

Step 6: On Resume

Return combined context for orchestrator to continue:

## Resuming Session

### Current State
[Content from context.md]

### Recent Decisions
[Last 5 entries from log.md]

### Next Actions
[From context.md Next Steps section]

Step 7: On Start Fresh

If user chooses fresh start:

  1. Move current session files to .tmp/archive/session-[timestamp]/
  2. Return empty context
  3. Orchestrator starts Phase 1

Implementation

### Check session exists

Read `.tmp/current/session/context.md`

If not found:
  Return "No previous session found. Starting fresh."

### Validate timestamp

Extract "Updated: YYYY-MM-DDTHH:MM:SS" from file
Calculate age in hours

If age > 24:
  Ask user: "Session is [X] hours old. Resume or start fresh?"

### Load and return

Read context.md fully
Read last 500 lines of log.md (covers ~10 entries)
Return combined content

Output Format

When session found:

Previous session found:
- Workflow: health-bugs
- Phase: 3/7 (Staged Fixing)
- Age: 2h 15m
- Last: Fixed auth validation in src/auth.ts

Resume? [Y/n]

When no session:

No previous session found. Starting fresh workflow.

When stale:

Found stale session (26 hours old):
- Workflow: health-security
- Phase: 2/5

[R]esume anyway, [A]rchive and start fresh, or [C]ancel?

Token Cost

  • Context read: ~100-150 tokens
  • Log read (last 10 entries): ~200-300 tokens
  • Total: ~300-450 tokens (only when resuming)
  • save-session-context — Save current state for later
  • load-project-context — Load project structure

Integration with Orchestrators

Health orchestrators should call this skill in Phase 0:

## Phase 0: Session Check

1. Invoke `resume-session` skill
2. If valid session:
   - Ask user: Resume or fresh?
   - If resume: Jump to saved phase
3. If no session:
   - Proceed to Phase 1
4. Create new session context