Skip to content
Back to skills

Ralph Resume Obsidian Owl Floe

ASecurity

Resume from saved session state after interruption or compaction. Recovers context from checkpoints and agent-memory.

  • 666 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 2, 2026
ai-agentsbash

Works with

  • claude code

Security analysis

A100/100

Scanned September 2, 2026

npx -y skills add majiayu000/claude-skill-registry --skill ralph-resume-obsidian-owl-floe --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Ralph Resume Obsidian Owl Floe?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Ralph Resume Obsidian Owl Floe
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-ralph-resume-obsidian-owl-floe/badge)](https://www.skillsdirectory.com/skills/majiayu000-ralph-resume-obsidian-owl-floe)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: ralph-resume
description: Resume from saved session state after interruption or compaction. Recovers context from checkpoints and agent-memory.
triggers:
  - /ralph.resume
  - resume session
  - recover session
---

# /ralph.resume [--session SESSION_ID]

Resume a workflow session from saved state.

## When to Use

- After fixing a BLOCKED pre-flight check
- After Claude Code context compaction
- After session interruption
- To continue from a specific checkpoint

## Process

1. **Check for Active Session**

   ```bash
   cat .ralph/session/current.json 2>/dev/null
   ```

   If exists, shows session details and asks to resume.

2. **Query Agent-Memory for Context** (if Cognee available)

   ```
   Search agent-memory for:
   "session context for {work_area}"
   ```

   Retrieve:
   - Prior decisions made during session
   - Task states and progress
   - Implementation context

3. **Load Local Checkpoint** (fallback)

   ```bash
   ls -la .ralph/session/checkpoints/
   # Find most recent checkpoint
   cat .ralph/session/checkpoints/{session_id}-{checkpoint_num}.json
   ```

4. **Restore State**

   - Load active task list from checkpoint
   - Restore Linear issue states
   - Resume from last completed subtask

5. **Run Pre-Flight Check**

   ```
   /ralph.preflight
   ```

   Must PASS before resuming work.

6. **Continue Workflow**

   Resume from the interrupted step.

## Output

```
RALPH WIGGUM SESSION RESUME
===========================

Session ID: session-2026-01-16-abc123
Started: 2026-01-16T10:30:00Z
Last checkpoint: 2026-01-16T11:45:00Z (iteration 8)

Prior Context Loaded:
  - 5 decisions recovered from agent-memory
  - 3 active tasks in progress
  - 12 subtasks completed

Active Tasks:
  T001 (auth): Iteration 8/15, subtask 3 of 5
  T003 (catalog): Iteration 5/15, subtask 2 of 4
  T005 (tests): COMPLETE

Pre-flight: PASS

Resuming from iteration 8...
```

## Recovery Without Cognee

If agent-memory is unavailable, local checkpoints are used:

```
RALPH WIGGUM SESSION RESUME
===========================

Session ID: session-2026-01-16-abc123

[WARN] Agent-memory unavailable - using local checkpoint only
       Some decision context may be missing

Checkpoint loaded: .ralph/session/checkpoints/session-2026-01-16-abc123-5.json
  - Last checkpoint: 2026-01-16T11:45:00Z
  - Active tasks: 3
  - Iteration: 8

Pre-flight: PASS (with warnings)
  [??] cognee: Using local buffer

Resuming from checkpoint...
```

## Session File Format

`.ralph/session/current.json`:

```json
{
  "session_id": "session-2026-01-16-abc123",
  "started_at": "2026-01-16T10:30:00Z",
  "work_area": "feature/ep001-auth",
  "epic": "EP001",
  "active_tasks": ["T001", "T003", "T005"],
  "last_checkpoint": 5,
  "status": "interrupted"
}
```

## Checkpoint Format

`.ralph/session/checkpoints/{session_id}-{num}.json`:

```json
{
  "checkpoint_id": "session-2026-01-16-abc123-5",
  "created_at": "2026-01-16T11:45:00Z",
  "iteration": 8,
  "active_tasks": {
    "T001": {"status": "in_progress", "subtask": 3, "passes": ["T001.1", "T001.2"]},
    "T003": {"status": "in_progress", "subtask": 2, "passes": ["T003.1"]},
    "T005": {"status": "complete", "subtask": 4, "passes": ["T005.1", "T005.2", "T005.3", "T005.4"]}
  },
  "decisions": [
    {"decision": "Use polling instead of sleep", "task_id": "T001", "timestamp": "..."},
    {"decision": "Add retry logic for catalog", "task_id": "T003", "timestamp": "..."}
  ]
}
```

## Manual Session Selection

If multiple sessions exist:

```bash
/ralph.resume --session session-2026-01-15-xyz789
```

Or list available sessions:

```bash
ls .ralph/session/checkpoints/ | cut -d'-' -f1-5 | sort -u
```

## Session Cleanup

After successful workflow completion:

```bash
# Archive completed session
mv .ralph/session/current.json .ralph/session/recovery/
rm .ralph/session/checkpoints/session-2026-01-16-abc123-*.json
```

## Configuration

From `.ralph/config.yaml`:

```yaml
resilience:
  checkpoints:
    session_ttl_hours: 48      # Retain session data
    max_checkpoints: 10        # Keep last 10 per session
  recovery:
    auto_save_on_block: true   # Save on BLOCKED signal
    auto_save_on_error: true   # Save on unexpected errors
```

## Related Commands

- `/ralph.preflight` - Check services before resume
- `/ralph.memory-sync` - Sync any buffered memories
- `/ralph.status` - Check current agent status

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.