/ralph-resume

namespace: aiwg
name: ralph-resume
platforms: [all]
description: Resume an interrupted agent loop from last checkpoint
commandHint:
  argumentHint: '[--max-iterations N] [--timeout M --interactive --guidance "text"]'
  allowedTools: "Task, Read, Write, Bash, Glob, Grep, TodoWrite, Edit"
  model: haiku
  category: automation
  orchestration: true
  modelRole: efficiency
  modelTier: economy

<!-- AIWG-SKILL-CALLOUT --> > **Skill access pattern (post-kernel-pivot, 2026.5+)** > > Skill names referenced in this document are AIWG skills, **not slash commands**. Most are not kernel-listed and cannot be invoked as `/skill-name` by the platform. Reach them via: > > ```bash > aiwg discover "<capability>" > aiwg show skill <name> --body > ``` > > Only kernel-listed skills (`aiwg-doctor`, `aiwg-refresh`, `aiwg-status`, `aiwg-help`, `use`, `steward`) are directly invokable as slash commands. See [skill-discovery rule](../../../addons/aiwg-utils/rules/skill-discovery.md).

Al Resume

Resume a paused or interrupted agent loop.

HITL authorization checkpoints

Before any confirmation below, read and apply [agent-loop authorization checkpoints](../../docs/hitl-authorization.md). Call `aiwg config authorization check --action-file "$ACTION_FILE" --target "$TARGET_REPO"` for host-classified effects. Only `auto-authorized` with a signed receipt skips that workflow prompt; retain existing prompts/refusals otherwise. Platform prompts are never bypassed. Limits, regression and threat overrides still require review.

Usage

/ralph-resume                       # Resume with existing settings
/ralph-resume --max-iterations 20   # Resume with higher iteration limit
/ralph-resume --timeout 120         # Resume with longer timeout

Parameters

--max-iterations N

Override the maximum iterations limit. Useful when loop stopped at limit but was making progress.

--timeout M

Override the timeout in minutes. Useful when loop timed out but task is close to completion.

Your Actions

Step 1: Load State

1. Read `.aiwg/ralph/current-loop.json` 2. Verify loop can be resumed (status != 'success', status != 'aborted') 3. Load iteration history and learnings

**If no resumable loop**:

No agent loop to resume.

Status: {status}

{If success}: Loop completed successfully. Start a new loop with /ralph
{If aborted}: Loop was aborted. Start fresh with /ralph
{If no state}: No loop found. Start with /ralph "task" --completion "criteria"

Step 2: Update Settings

Apply any parameter overrides:

• Update `maxIterations` if --max-iterations provided
• Update `timeoutMinutes` if --timeout provided
• Reset timeout start time for extended timeout

Step 3: Resume Execution

Continue the agent loop pattern:

1. Display resume status:

Resuming Agent Loop

Task: {task}
Completion: {completion}
Previous iterations: {N}
Remaining iterations: {max - N}

Last result: {lastResult}
Learnings so far: {learnings}

Continuing from iteration {N+1}...

2. Execute next iteration with accumulated learnings 3. Follow standard agent loop verification 4. Continue until success or new limits reached

Step 4: Handle Completion

Same as `ralph` - generate completion report on success or limit.

Resume Context

When resuming, include in the task context:

## Agent Loop Resume Context

**Original Task**: {task}
**Completion Criteria**: {completion}

**Previous Iterations**: {N}
**Accumulated Learnings**:
{for each iteration}
- Iteration {i}: {action} -> {result}. Learned: {learnings}
{end for}

**Current State**:
- Last attempt: {lastResult}
- Key insight: {most recent learning}

**Your Goal**:
Continue iterating from iteration {N+1}.
Apply learnings from previous iterations.
Verify against completion criteria after each attempt.

Error Handling

**Loop completed successfully**:

This agent loop already completed successfully.

Final status: SUCCESS
Iterations: {N}
Report: .aiwg/ralph/completion-{timestamp}.md

To run again, start a new loop:
  /ralph "task" --completion "criteria"

**Loop was aborted**:

This agent loop was aborted and cannot be resumed.

To start fresh with the same task:
  /ralph "{original task}" --completion "{original completion}"

**State corrupted**:

Agent loop state is corrupted or incomplete.

Options:
1. Start fresh: /ralph "task" --completion "criteria"
2. Clean up: rm -rf .aiwg/ralph/ then start new loop

Example Scenarios

Max Iterations Override

Previous loop stopped at iteration 10:

/ralph-resume --max-iterations 20

Continues with 10 more iterations available.

Timeout Override

Previous loop timed out at 60 minutes:

/ralph-resume --timeout 120

Continues with fresh 120-minute timeout.

Simple Resume

Loop interrupted (network, restart, etc.):

/ralph-resume

Continues from last checkpoint with original settings.

Related

• `ralph-status` - Check what state the loop is in
• `ralph-abort` - Stop instead of resume
• `ralph` - Start new loop

References

• @$AIWG_ROOT/agentic/code/addons/agent-loop/README.md — Ralph addon overview and loop executor documentation
• @$AIWG_ROOT/agentic/code/addons/aiwg-utils/rules/vague-discretion.md — Loop termination and iteration limit rules
• @$AIWG_ROOT/docs/cli-reference.md — CLI reference for ralph-resume and related commands
• @$AIWG_ROOT/agentic/code/addons/aiwg-utils/rules/instruction-comprehension.md — Re-reading original task instructions on resume