Work with the archive

Fetch all the data.

The website is one view of the collection. The repository is the archive itself, designed to be cloned, searched, transformed, and analyzed by people or agents.

Clone the repository

One command, then local files.

The public repository contains the archive records and prompt artifacts cleared for redistribution.

Open on GitHub ↗

git clone https://github.com/wintercarver/prompts-for-progress.git
cd prompts-for-progress

Repository map

Start with the generated corpus.

The Markdown remains the editable source of truth. The JSONL and JSON indexes are generated views intended for analysis.

01prompts/corpus.jsonl

One JSON object per locally preserved prompt artifact. This is the simplest starting point for aggregate analysis.

02prompts/corpus-index.json

One summary per prompt manifest, including completeness, publication status, artifact count, and known gaps.

03records/*.md

Structured record metadata in JSON frontmatter, followed by short editorial context in Markdown.

04prompts/sources/

The underlying prompt files, code, notebooks, configurations, and normalized transcripts.

Command line

Inspect the corpus with jq.

# Count prompt artifacts by completeness
jq -s 'group_by(.completeness) | map({completeness: .[0].completeness, artifacts: length})' prompts/corpus.jsonl

# List exact prompt artifacts
jq -r 'select(.completeness == "exact") | [.recordId, .sourcePath] | @tsv' prompts/corpus.jsonl

Python

Load every artifact.

import json
from pathlib import Path

path = Path("prompts/corpus.jsonl")
rows = [json.loads(line) for line in path.read_text().splitlines()]

exact = [row for row in rows if row["completeness"] == "exact"]
print(len(rows), len(exact))

Check rights metadata before republishing prompt text.

Each corpus row includes a publication status. The public corpus contains only material marked approved. Known prompt artifacts awaiting rights review are omitted while their original sources remain linked from archive records.