makememe — a meme CLI your coding agent can drive

A tiny, zero-dependency meme generator for coding agents and CI. No API key, no signup, stdlib-only — it wraps the free memegen.link API and hands back a public image URL you can embed anywhere with ![meme](url).

Quick start

Give your coding agent a meme button. Two commands, then just ask:

uv tool install makememe          # or: pipx install makememe  /  pip install makememe
meme --install-skill              # installs the bundled Claude Code skill

Restart Claude Code and talk to it normally:

you:   make a "this is fine" meme about prod being down
Claude: → https://api.memegen.link/images/fine/prod_is_down/this_is_fine.png

The skill teaches the agent to pick the template that fits the joke (not just drake), write the caption, and keep any names you mention — then hand back a URL that renders inline in a PR, Slack, or anywhere. (Codex and other agents work out of the box via meme --help / meme --list — no skill needed.)

meme drake "manual deploys" "ci/cd"                       # saves a PNG to a temp folder
meme -t fine "prod is down" "this is fine" --print-url    # or just print a shareable URL
meme --list                                               # browse all template ids

Next: meme your CI on every PR with a copy-paste GitHub Action.

Install / upgrade

# install (pick one)
uv tool install makememe          # recommended — isolated, puts `meme` on PATH
pipx install makememe
python3 -m pip install makememe

# install the bundled Claude Code skill, then restart Claude Code
meme --install-skill              # ~/.claude/skills/meme/ (all projects)
meme --install-skill --project    # ./.claude/skills/meme/ (this repo only)

Upgrading:

uv tool upgrade makememe          # pipx: pipx upgrade makememe | pip: pip install -U makememe
meme --install-skill              # ⚠️ re-run this so the skill updates too

⚠️ Upgrading the package does not refresh the skill already copied into ~/.claude/skills/meme/. Re-run meme --install-skill after every upgrade (then restart Claude Code), or the agent keeps using the old skill.

Skip the approval prompt (optional). Claude Code asks once before running meme. To never be prompted, add this to the permissions → allow list in your Claude Code settings.json:

"Bash(meme:*)"

Verify:

meme --version                    # CLI version
meme --list                       # CLI runs and reaches templates
ls ~/.claude/skills/meme/SKILL.md # skill is installed

Usage

meme <template> "top line" "bottom line" [-o out.png]

Examples

meme drake "manual deploys" "ci/cd"
meme same "after I sold" "if I held" "same picture"
meme --bg https://example.com/pic.png "_" "DODGED"
meme regret "SOLD @ 620" "NOW 780 (+26%)" --print-url
meme --list
meme --list --json

Integrations

For sharing (CI, chat, comments) you usually want the public URL, not a local file — --print-url returns a permanent memegen.link URL you can embed anywhere with ![meme](url). No download, no image hosting.

Meme your CI

Drop this in .github/workflows/pr-meme.yml, change the one test line to your command, and every PR gets a success / "this is fine" meme based on whether tests passed. That's the whole setup — no secrets (GITHUB_TOKEN is built in):

name: pr-meme
on: pull_request
permissions:
  pull-requests: write          # to post the comment
jobs:
  meme:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Run tests
        id: tests
        run: echo "replace this with your test command (e.g. pytest, npm test)"
      - name: Comment a meme with the result
        if: always()            # run even when tests fail
        run: |
          pip install --quiet makememe
          if [ "${{ steps.tests.outcome }}" = "success" ]; then
            url=$(meme -t success "tests" "passed" --print-url)
          else
            url=$(meme -t fine "tests failed" "this is fine" --print-url)
          fi
          gh pr comment "${{ github.event.pull_request.number }}" --body "![meme]($url)"
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

(Also saved at examples/pr-meme.yml.)

Meme your release

Celebrate a tag with a meme in the release notes or a Slack announcement — just feed it the version:

ver=$(git describe --tags --abbrev=0)
url=$(meme -t success "shipped" "$ver" --print-url)
# drop $url into your GitHub Release body, changelog, or a Slack post

Or just ask your agent: "create a meme about this release" — it reads the version/changelog and picks a fitting template.

Slack/Discord: post a --print-url URL to a channel; it auto-unfurls into a preview.

url=$(meme -t success "build" "passed" --print-url)   # -> public URL, embed it anywhere

Text that starts with -

If a caption line begins with - (e.g. "-26%"), put -- before your lines so it isn't parsed as a flag:

meme regret --json -- "-26%" "WHY"

(Put flags like --json/-o before the --.)

For agents (Claude Code / Codex / scripts)

The tool is designed to be parsed:

• Plain mode prints only the output path to stdout (status goes to stderr), so path=$(meme drake "a" "b") just works.

• --json mode prints a single JSON object:

  { "path": "/tmp/makememe/meme-ab12cd.png", "bytes": 12345, "url": "https://api.memegen.link/..." }
  

  --list --json returns the template catalog as JSON; --print-url --json returns {"url": "..."}; failures return {"error": "...", "url": "..."} with a non-zero exit code.

Typical agent flow:

meme --list --json                       # discover template ids
meme drake "old way" "new way" --json    # generate, capture the path

Claude Code skill

The package bundles a Claude Code skill. After installing, run one command to make Claude Code auto-discover the tool:

meme --install-skill              # installs into ~/.claude/skills/meme/ (all projects)
meme --install-skill --project    # or into ./.claude/skills/meme/ (this repo only)

Restart Claude Code, then just talk to it. You don't name templates or flags — the skill picks the template that fits the joke, writes the caption, and keeps any names you mention. Things you can say:

It's conversational — "funnier", "use the two-buttons template instead", or "now make the bottom line shorter" all work as follow-ups.

(Other agents like Codex don't use this skill format — they discover everything through meme --help and meme --list, which already works out of the box.)

Robustness

The CLI is built to fail gracefully, never with a raw traceback:

• Network errors, dead hosts, bad template ids (404), oversized text (414), and non-image backgrounds (415) all exit non-zero with a one-line message — and no partial/garbage file is written.
• Ctrl-C exits cleanly (code 130); piping into head etc. won't spew a BrokenPipeError.
• Arbitrary text — emoji, CJK, % # & / ? " \, tabs, control chars, 10k-char lines — is escaped safely.

Run the test suite (stdlib only, no network needed):

python -m unittest discover -s tests

How it works

It builds a memegen.link URL from your template + text (handling all the fiddly path-segment escaping — spaces, _, -, ?, /, %, etc.), downloads the image, and saves it. That's the whole trick.

License

MIT