gemini-image-recognize-install

name	gemini-image-recognize-install
description	Make the gemini-image-recognize command available, installing it only if it is missing. Use when another skill reports that `gemini-image-recognize` is not on PATH, or when the user asks to install, update or upgrade the ideamans Gemini image recognition CLI. Prefers an already-installed binary, then the latest GitHub release, then a build from source with go install.
license	MIT
compatibility	Requires curl (or wget) and tar to install from a release, or a Go toolchain for the source fallback. Standalone — does not need gemini-image-recognize to be present already. Installs from the public repository github.com/ideamans/gemini-image-recognize-cli, so no GitHub authentication is needed.
allowed-tools	Bash(curl:*) Bash(wget:*) Bash(tar:*) Bash(unzip:*) Bash(go:*) Bash(uname:*) Bash(command:*) Bash(which:*) Bash(mkdir:*) Bash(mv:*) Bash(cp:*) Bash(rm:*) Bash(chmod:*) Bash(ls:*) Bash(test:*) Bash(echo:*) Read

gemini-image-recognize-install

Make the gemini-image-recognize command usable, doing the least work that achieves it.

Route 1 — an existing installation on PATH

command -v gemini-image-recognize && gemini-image-recognize --version

If that resolves, use it and stop here. Do not check for a newer release — it costs an API call and the user did not ask for an upgrade.

Two checks before trusting the hit:

• It is the right tool. gemini-image-recognize llm | head -1 must read # gemini-image-recognize — Gemini image recognition CLI (reference for AI agents). If something else owns the name, tell the user and use an explicit path rather than shadowing theirs.
• It is recent enough. If gemini-image-recognize llm is not a known command, the binary predates the embedded reference. Say so and continue to route 2 to upgrade it.

Continue past this section only when the command is missing, is the wrong tool, is too old, or the user explicitly asked to update.

Route 2 — the latest GitHub release

The repository is public, so no authentication is needed.

VERSION=$(curl -fsSL https://api.github.com/repos/ideamans/gemini-image-recognize-cli/releases/latest \
  | grep '"tag_name"' | head -1 | cut -d'"' -f4)   # e.g. v0.1.0

The archive is named after the goreleaser project, not the repository — gemini-image-recognize, without the -cli suffix:

gemini-image-recognize_<version-without-v>_<os>_<arch>.tar.gz

<os> is darwin, linux or windows (lowercase); <arch> is amd64 or arm64, so uname -m reporting x86_64 maps to amd64. Windows ships a .zip.

OS=$(uname -s | tr '[:upper:]' '[:lower:]')             # darwin | linux
ARCH=$(uname -m); [ "$ARCH" = "x86_64" ] && ARCH=amd64  # amd64 | arm64
curl -fsSL -o /tmp/gemini-image-recognize.tar.gz \
  "https://github.com/ideamans/gemini-image-recognize-cli/releases/download/${VERSION}/gemini-image-recognize_${VERSION#v}_${OS}_${ARCH}.tar.gz"

If the download 404s, list the actual assets on the release page rather than retrying variations.

Install onto PATH

tar -xzf /tmp/gemini-image-recognize.tar.gz -C /tmp
mkdir -p ~/.local/bin && mv /tmp/gemini-image-recognize ~/.local/bin/ \
  && chmod +x ~/.local/bin/gemini-image-recognize

Prefer the first writable directory already on PATH — ~/.local/bin, then /usr/local/bin. Two things not to do on your own initiative:

• If nothing on PATH is writable, leave the binary in /tmp, print the exact sudo mv command and let the user run it. Do not run sudo yourself.
• If ~/.local/bin is not on PATH, give the user the line to add to their shell profile. Do not edit the profile for them.

Route 3 — build from source

Needs a Go toolchain and compiles rather than downloads, so it is the last resort — but it covers platforms the release assets miss. Note the /cmd/gemini-image-recognize suffix; installing the module root would not build anything.

go install github.com/ideamans/gemini-image-recognize-cli/cmd/gemini-image-recognize@latest

The binary lands in $(go env GOPATH)/bin and is named gemini-image-recognize, the same as the release archive's.

Verify

gemini-image-recognize --version
gemini-image-recognize llm | head -5

Report which route was taken, the version and the install path.

Then say what is still needed: the CLI cannot describe anything without a Gemini API key. Tell the user now rather than letting the first real command fail — they need a key from https://aistudio.google.com/apikey, supplied either as

export GEMINI_API_KEY=<key>

or stored once with gemini-image-recognize configure --key-file <path>. Verify with gemini-image-recognize auth status, and never ask them to paste the key into the conversation.