Screenshot Stitcher

Use when the task is to stitch multiple vertically scrolling iPhone screenshots into a single long image with the local screenshot-stitcher CLI. Best for sam...

Security Scan

The skill's requirements and instructions are consistent with a local CLI tool for stitching iPhone screenshots; it asks for no credentials and has no surprising install steps, but installing a third‑party PyPI package or running an unchecked repo entrypoint has the normal risks of executing remote code.

Purpose & Capability

Name, description, and runtime instructions all align: the skill expects a local screenshot-stitcher CLI (PyPI package or repo entrypoint) and only describes reading/writing image files needed for the task.

Instruction Scope

Instructions are scoped to running the local CLI on image files and reporting results. One caution: the fallback to 'python3 main.py' in a cloned checkout will execute arbitrary code in the current directory — that's expected as a developer fallback but increases risk if the checkout is untrusted. The privacy claim (local processing only) is reasonable but depends on the actual package content, which isn't provided here.

Install Mechanism

The SKILL.md recommends 'pip install screenshot-stitcher' (PyPI). Installing from PyPI is typical and proportionate, but installing any third-party package runs remote code on the machine. There is no bundled install spec in the skill itself (instruction-only).

Credentials

No environment variables, credentials, or config paths are requested — permissions requested are minimal and appropriate for a local CLI image tool.

Persistence & Privilege

Skill does not request always:true, does not modify other skills, and does not request persistent privileges. It is user-invocable and may be invoked autonomously by the agent (platform default).

Assessment

This skill appears coherent and minimal: it expects you to install and run a local PyPI CLI to stitch screenshots and does not request credentials. Before installing or running it, consider: 1) only install the package in a controlled Python environment (virtualenv) and verify the package/source if you don't trust it; 2) avoid running 'python3 main.py' from an untrusted checkout (that runs arbitrary code); 3) if you need stronger guarantees, review the package source on PyPI/GitHub or prefer an offline/air-gapped process. If you plan to let an autonomous agent run these commands, be aware that installing a package or executing a local script gives that agent the ability to run code on your machine.

Use this skill when the user wants to combine multiple iPhone screenshots into one long image.

The CLI contract is intentionally small, so the same workflow ports cleanly into Claude, Codex, OpenClaw, Hermes, and similar agent wrappers that can invoke shell commands. The tool is OpenCV/NumPy-driven local image processing, not AI image generation.

Privacy

All image processing happens locally in the user's Python process. The CLI reads screenshots from disk, writes the stitched output back to disk, and does not upload screenshots, filenames, or image metadata to any external service.

Install The CLI

Prefer the published PyPI package:

pip install screenshot-stitcher
screenshot-stitcher --help

If the machine has multiple Python environments, install into the environment the agent will use:

python -m pip install screenshot-stitcher
screenshot-stitcher --help

If the package is already installed, skip installation and use the CLI directly.

Workflow

1. Confirm the inputs are vertically scrolling screenshots from the same device width.
2. Preserve the user's input order. Do not reorder by timestamp.
3. Use the installed CLI:

screenshot-stitcher img1.png img2.png img3.png -o output.png

4. If the package cannot be installed but the agent is inside a cloned checkout, run the repo entrypoint instead:

python3 main.py img1.png img2.png img3.png -o output.png

5. After writing the output file, report the path and mention any low-confidence overlaps printed by the CLI.

Useful Flags

• --no-navbar: page has no navigation bar
• --no-tabbar: page has no bottom tab bar
• --top-crop N: manual top crop override
• --bottom-crop N: manual bottom crop override
• --x-margin N: ignore more left/right edge pixels during matching
• --template-height N: adjust the template height used for overlap matching
• --threshold V: raise or lower overlap acceptance sensitivity

Heuristics

• Start with the default command before tuning flags.
• If the page has sticky bars, floating controls, or scroll indicators, try a larger --x-margin.
• If the top chrome changes during scrolling, try --no-navbar or a custom --top-crop.
• If overlap confidence is low but the output is visually correct, report that explicitly instead of pretending the score is strong.
• If the input screenshots are out of order or have different widths, stop and surface that clearly.