Quality

/archive

Use when planning a sprint transition, archiving completed work, or resetting the PRD for a fresh start. Archives completed PRD stories to docs.local/. Covers archive, cleanup, sprint transition, reset PRD. NOT for: deleting stories mid-sprint (use prd-manager), viewing PRD status (use ralph-status).

$ golems-cli skills install archive

Experimental

100% best pass rate

8 assertions

2 evals

fixtures

2 workflows

Updated 2 weeks ago

Archive completed stories to docs.local/prd-archive/ with full history preservation. Optionally reset the working PRD for a fresh start.

Available Scripts

Run these directly - standalone, no ralph.zsh dependency:

Script	Purpose	Usage
`scripts/archive-snapshot.sh`	Create archive snapshot	`bash ~/.claude/commands/archive/scripts/archive-snapshot.sh`
`scripts/cleanup-completed.sh`	Archive + remove completed	`bash ~/.claude/commands/archive/scripts/cleanup-completed.sh`

Quick Actions

What you want to do	Workflow
Archive current PRD state (snapshot)	workflows/archive-snapshot.md
Remove completed stories (cleanup)	workflows/cleanup-completed.md

What Gets Archived

Each archive creates a timestamped directory:

docs.local/prd-archive/20260123-153000/
├── index.json           # PRD index at time of archive
├── stories/             # All story files (completed + pending)
│   ├── US-001.json
│   ├── US-002.json
│   └── ...
└── progress.txt         # Ralph progress log

Full SKILL.md source — includes LLM directives, anti-patterns, and technical instructions stripped from the Overview tab.

Archive completed stories to docs.local/prd-archive/ with full history preservation. Optionally reset the working PRD for a fresh start.

Available Scripts

Run these directly - standalone, no ralph.zsh dependency:

Script	Purpose	Usage
`scripts/archive-snapshot.sh`	Create archive snapshot	`bash ~/.claude/commands/archive/scripts/archive-snapshot.sh`
`scripts/cleanup-completed.sh`	Archive + remove completed	`bash ~/.claude/commands/archive/scripts/cleanup-completed.sh`

Quick Actions

What you want to do	Workflow
Archive current PRD state (snapshot)	workflows/archive-snapshot.md
Remove completed stories (cleanup)	workflows/cleanup-completed.md

What Gets Archived

Each archive creates a timestamped directory:

docs.local/prd-archive/20260123-153000/
├── index.json           # PRD index at time of archive
├── stories/             # All story files (completed + pending)
│   ├── US-001.json
│   ├── US-002.json
│   └── ...
└── progress.txt         # Ralph progress log

Cleanup Behavior

When cleanup is triggered (via --clean or confirming prompt):

Completed stories deleted - All *.json files where passes=true
Index reset - Stats reset, pending array rebuilt from remaining stories
Progress cleared - progress.txt replaced with fresh start header

History is always preserved - The archive contains full state before cleanup.

Safety Notes

Always archives first - Cleanup only happens after successful archive
Blocked stories preserved - Stories in blocked array are kept
Pending stories preserved - Only completed (passes=true) stories removed
Reversible - Full state available in docs.local/prd-archive/

Experimental

Best Pass Rate

100%

Opus 4.6

Assertions

6 models tested

Avg Cost / Run

$0.0914

across models

Fastest (p50)

2.3s

Sonnet 4.6

Behavior Evals

Phase 2 baseline — skill quality on Claude

Behavior Baseline

Opus 4.6

100%8/8

●

Haiku 4.5

100%8/8

●

Sonnet 4.6

88%7/8

◒

Adapter Evals

Phase 2C — cross-AI portability

Adapter Portability

Cursor

100%8/8

●

Codex

75%6/8

◒

Gemini 2.5

75%6/8

◒

Assertion	Opus 4.6	Sonnet 4.6	Haiku 4.5	Codex	Gemini 2.5	Cursor	Consensus
creates-timestamped-archive-directory							5/6
archives-before-cleanup							6/6
preserves-pending-and-blocked							5/6
resets-prd-index-after-cleanup							4/6
uses-archive-scripts							6/6
creates-snapshot-without-cleanup							6/6
archive-contains-full-state							6/6
no-stories-deleted							5/6

Token Usage

Opus 4.6

4,761

Sonnet 4.6

13,143

Haiku 4.5

3,682

Codex

6,146

Gemini 2.5

4,958

Cursor

11,638

Input tokensOutput tokens

Cost per Run

Opus 4.6

$0.1943

Sonnet 4.6

$0.1296

Haiku 4.5

$0.0029

Codex

$0.0660

Gemini 2.5

$0.0322

Cursor

$0.1231

Model	Input Tokens	Output Tokens	Cost / Run	Cost / 1K Runs
Opus 4.6	2,713	2,048	$0.1943	$194.30
Sonnet 4.6	5,626	7,517	$0.1296	$129.60
Haiku 4.5	1,704	1,978	$0.0029	$2.90
Codex	3,792	2,354	$0.0660	$66.00
Gemini 2.5	2,318	2,640	$0.0322	$32.20
Cursor	5,262	6,376	$0.1231	$123.10

Response Time (p50)

Sonnet 4.6

2.3s

Gemini 2.5

2.4s

Haiku 4.5

2.8s

Codex

5.3s

Cursor

5.7s

Opus 4.6

9.4s

Response Time (p95)

Sonnet 4.6

4.5s

Gemini 2.5

4.6s

Haiku 4.5

5.5s

Codex

9.2s

Cursor

9.4s

Opus 4.6

15.9s

Model	p50	p95	Overhead
Opus 4.6	9.4s	15.9s	+69%
Sonnet 4.6	2.3s	4.5s	+92%
Haiku 4.5	2.8s	5.5s	+94%
Codex	5.3s	9.2s	+74%
Gemini 2.5	2.4s	4.6s	+91%
Cursor	5.7s	9.4s	+64%

Last evaluated: 2026-03-12 · Data is generated from skill assertions (real cross-model benchmarks coming soon)

Workflows

/archive:cleanup-completed/archive:archive-snapshot