Skip to content

v0.4.0

Choose a tag to compare

@github-actions github-actions released this 15 Oct 19:56
· 7 commits to main since this release

Fewer steps for a coding agent than genmedia (fal's CLI), from a fresh machine to a saved file.

Added

  • opensume init writes the agent page where coding agents read instructions: a marked block in AGENTS.md, .claude/skills/opensume/SKILL.md, and .cursor/rules/opensume.mdc when the project has a .cursor folder. Running it again refreshes them. It touches nothing else and makes no network calls. --targets picks which files to write.
  • opensume models --brief: one line per model with its price per image or second, its inputs and its limits, cheapest first in each kind. With --json it is about 12 KB, against 85 KB for the full catalog.
  • opensume agent ends with that list, from the public catalog with no key needed, so one command gives an agent the setup and the choices. Its JSON has models.
  • The key is also read from SUME_API_KEY in ./.env in the current directory, so there is no setup command. Precedence: flag, then the environment, then ./.env, then the saved key. The API host is never read from .env.
  • OPENSUME_MAX_USD caps every run when --max-usd is not given. The refusal names which cap it hit.
  • install.sh links opensume into ~/.local/bin when that folder is already on PATH, so the command works in the same shell. Otherwise it prints the full path to use until the shell is reloaded. The release workflow checks this on macOS and Linux.

Changed

  • The agent page is one source for opensume agent, the AGENTS.md block, the Claude skill, the Cursor rule and this repo's skills/opensume/SKILL.md. It is shorter and leads with the three commands. Tests keep the generated files equal to it.
  • The install scripts and the no-key error say a person makes the key, and that ./.env works.