# Fetch LLM Apps

> Workflow for updating the popular LLM applications pool (section/x_llm_apps.md) using fetch_llm_apps.py. Covers full refresh, alternate exports, topic tuning, and common pitfalls. USE FOR: Refreshing the ranked GitHub applications list linked from applications.md. DO NOT USE FOR: Hand-curating application entries inside applications.md or adding GitHub star badges to the generated file.

- Skill: `gabrielmoreira/fetch-llm-apps` (Agent Skill)
- Install (CLI): `npx skillmds@latest add gabrielmoreira/fetch-llm-apps`
- Raw SKILL.md: https://api.skillmd.com/api/skills/gabrielmoreira/fetch-llm-apps/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: gabrielmoreira (https://skillmd.com/u/gabrielmoreira)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/gabrielmoreira/fetch-llm-apps

---


## Overview

The pool file `section/x_llm_apps.md` is a generated ranked list of GitHub repositories related to LLM apps, agents, chat UIs, workflow builders, and similar application-layer projects.

It is generated by `code/fetch_llm_apps.py` using GitHub topic search, deduplicated across multiple topics, and sorted by GitHub star count descending.

The section `#### Popular LLM Applications (GitHub Stars >= 1000)` in `section/applications.md` links to this generated file with a one-line description only. Do not paste generated entries directly into `applications.md`.

---

## Script Reference

**Script:** `code/fetch_llm_apps.py`  
**Python env:** `.venv\Scripts\python.exe`

### Key CLI Arguments

| Argument | Default | Purpose |
|----------|---------|---------|
| `--output` | `section/x_llm_apps.md` | Output file path. Extension controls format: `.md`, `.json`, `.csv` |
| `--min-stars` | `1000` | Minimum GitHub star threshold |
| `--topics` | curated list | GitHub topics to query and merge |
| `--token` | `GITHUB_TOKEN` env var | GitHub PAT for higher rate limits |
| `--show` | `30` | Number of repos printed to console |
| `--timeout` | `20` | Per-request timeout in seconds |
| `--max-retries` | `4` | Max retries per request |
| `--backoff` | `1.0` | Initial retry backoff |
| `--sleep` | `1.0` | Delay between successful page requests |
| `--include-archived` | off | Include archived repositories |
| `--append` | off | Merge new results with an existing output file, then re-sort by stars |

---

## Workflow

### 1. Full refresh of the markdown pool

Use this for the normal update path.

```powershell
.venv\Scripts\python.exe code/fetch_llm_apps.py
```

- Rewrites `section/x_llm_apps.md`.
- Uses `--min-stars 1000` by default to match the section title.
- Excludes archived repos unless `--include-archived` is passed.
- Writes compact numbered entries instead of badge-heavy markdown.

### 2. Authenticated refresh to avoid GitHub API limits

```powershell
.venv\Scripts\python.exe code/fetch_llm_apps.py --token $env:GITHUB_TOKEN
```

Use a GitHub PAT when doing a full refresh across many topics. Unauthenticated search is heavily rate-limited.

### 3. Generate a stricter ranking

```powershell
.venv\Scripts\python.exe code/fetch_llm_apps.py --min-stars 2000
```

Use this when you want a tighter list. If you change the threshold materially, update the descriptive text in `section/applications.md` so the label stays truthful.

### 4. Export raw data for review

```powershell
.venv\Scripts\python.exe code/fetch_llm_apps.py --output files/x_llm_apps.json
.venv\Scripts\python.exe code/fetch_llm_apps.py --output files/x_llm_apps.csv
```

Use JSON or CSV when you want to inspect or post-process the ranked repo pool before regenerating markdown.

### 5. Tune the topic set

```powershell
.venv\Scripts\python.exe code/fetch_llm_apps.py --topics llm agent rag chatbot ai-workflow
```

- Topics are GitHub repository topics, not free-text queries.
- Results are deduplicated by `full_name` after all topic passes complete.
- If the topic set changes substantially, regenerate the markdown pool rather than editing it by hand.

### 6. Topic-specific doc update

If the output is intentionally narrowed to a subset such as `gemini claude azure-openai copilot assistant`, keep the ranked entries as generated, then update the document metadata and the linking description in `section/applications.md` to reflect the narrowed scope.

### 7. Add topics without discarding the current pool

```powershell
.venv\Scripts\python.exe code/fetch_llm_apps.py `
	--append `
	--topics llm agent rag chatbot ai-workflow
```

`--append` parses the existing compact entries, merges newly fetched repositories by `full_name`, and rewrites the file sorted by star count. Use it when expanding coverage; use the normal full refresh when the default topic set changes substantially.

---

## Output Format

Each entry in `section/x_llm_apps.md` follows this compact format:

```markdown
1. [owner/repo](https://github.com/owner/repo): Short GitHub description. [Mon YYYY] (⭐ 12,345)
```

- Entries are sorted by GitHub stars descending.
- The date is the repository creation month, formatted as `[Mon YYYY]`.
- The star count is static text in parentheses using the star symbol.
- Do **not** append realtime shields or badges in this generated file.

The file header includes:

- UTC generated timestamp
- GitHub Search API source note
- searched topic list
- total repository count

---

## Common Pitfalls

1. **Using the wrong output target:** The generated ranking belongs in `section/x_llm_apps.md`, not inline inside `section/applications.md`.

2. **Adding GitHub star badges:** This file intentionally uses static text like `(⭐ 12,345)`. Do not run `add_github_stars.py` on it.

3. **Forgetting the star threshold contract:** The linked section title says `≥1000`. If you generate with a different threshold, either restore `1000` or update the section label and description.

4. **Unauthenticated rate limits:** Full runs over many topics can stall or fail without a token. Prefer `GITHUB_TOKEN` for routine refreshes.

5. **Assuming GitHub topics are comprehensive:** Some strong repos do not declare useful topics and may be missed. Expand `--topics` or curate separately if coverage is insufficient.

6. **Archived repos polluting the ranking:** Archived repos are excluded by default. Only include them deliberately.

7. **Hand-editing generated entries:** Manual changes will be lost on the next run. Adjust the script inputs or post-process separately instead.