MacOS Gemini image gen via OpenRouter for OpenClaw agents. Auto-wait, gen, send image.

This skill appears to be trying to do what it says (generate images via OpenRouter) but contains several inconsistencies and bugs you should resolve before installing or running it with real secrets: - Do not provide credentials until fixed: OPENROUTER_API_KEY is required by the script; TELEGRAM_BOT_TOKEN is listed but unused — clarify whether the skill actually needs a Telegram token. The package metadata should declare required env vars correctly. - Fix the runtime mismatch: SKILL.md expects ../output.json but scripts/gen.py prints JSON to stdout and writes image files to /tmp. Either change SKILL.md to capture stdout into ../output.json or modify gen.py to write ../output.json. As-is, the workflow will fail. - Fix the code bug: scripts/gen.py references item.get(b64_json) (an undefined name) when constructing a data URL; that will raise an error if that branch runs. Review and correct that logic. - Review external fetch behavior: the script downloads arbitrary URLs returned by OpenRouter and writes them to /tmp. This is expected for an image generator but entails network fetches and writing data to disk — run in a sandbox and limit exposure if you have strict security requirements. - Audit OpenRouter usage: the script posts to https://openrouter.ai; confirm you trust that service and that the API key scope is limited. Consider rate limits and log handling. If you need this skill, ask the publisher to (1) update metadata to declare OPENROUTER_API_KEY, (2) remove or justify TELEGRAM_BOT_TOKEN, (3) correct SKILL.md to match gen.py behavior (or vice-versa), and (4) fix the b64_json bug. Until these issues are addressed, treat the skill as untrusted and run only in an isolated environment.

Type: OpenClaw Skill Name: hana-image-gen-macos Version: 1.0.0 The skill contains a shell injection vulnerability in the `SKILL.md` workflow, which passes the user-controlled `$prompt` variable directly into a shell command (`python3 gen.py "$prompt"`) without sanitization. While the `scripts/gen.py` file appears to be a legitimate implementation for interacting with the OpenRouter API (https://openrouter.ai), the architectural flaw in the workflow allows for potential Command Injection. There is also a functional discrepancy where the workflow attempts to read from `../output.json` while the script only outputs results to stdout.