Skip to content

Releases

agentbuffs 0.5.0

Choose a tag to compare

@github-actions github-actions released this 06 Nov 10:28

Changed

  • "Show the changes first" and --dry-run print the full text of the skill file, not only its size and checksum. The skill holds instructions your agent follows, so you can now read them before you approve.
  • The skill preview shows control characters and invisible characters, such as escape codes and zero-width marks, as <U+XXXX> codes. The file is still written exactly as downloaded.

agentbuffs 0.4.1

Choose a tag to compare

@github-actions github-actions released this 01 Nov 10:51

Fixed

  • repository.url in package.json uses the GitHub organization name in lower case, agentbuffs. npm compares it with the provenance of the build and refused the 0.4.0 publish. 0.4.0 has a git tag but was never on npm. 0.4.1 has the same changes.

agentbuffs 0.3.2

Choose a tag to compare

@AltanS AltanS released this 01 Nov 09:48

Fixed

  • The --help example names opencode, so it no longer reads as the full list of agents.
  • --dry-run alone, without a terminal, lists every agent slug and exits 0. A plain run without --client lists them in its error.

agentbuffs 0.3.1

Choose a tag to compare

@AltanS AltanS released this 01 Nov 09:48

Fixed

  • --yes never asks. Without --what it installs both parts. Without --client it uses the agents it finds, and exits 2 with the known slugs when it finds none.
  • OpenCode follows XDG_CONFIG_HOME when it is an absolute path.
  • An XDG_CONFIG_HOME that points outside your home folder and the project is refused. The tool says so and leaves that agent for you to set up by hand.

agentbuffs 0.3.0

Choose a tag to compare

@AltanS AltanS released this 01 Nov 09:48

Added

  • In a terminal, the installer runs as a full-screen wizard with one screen per step: agents, parts, plan, install.
  • Long lists and the list of changes page inside the prompt. Every screen fits 80 by 24.
  • --no-fullscreen prints line by line instead.

Changed

  • Every way out (done, cancel, Ctrl+C, SIGTERM, an error) restores the terminal and leaves a short summary in the scrollback.
  • With --yes, NO_COLOR, TERM=dumb, no terminal, or a terminal smaller than 60 by 18, the output stays line by line.

agentbuffs 0.2.0

Choose a tag to compare

@AltanS AltanS released this 01 Nov 09:48

Added

  • A banner with the flame mark, in 24-bit, 256 or 16 colors, with a shape-only form for NO_COLOR.
  • --verbose shows every change before it asks.

Changed

  • The plan is one line per agent. "Show the changes first" and --dry-run list the paths, entries and commands.
  • One result line per agent, with a check, a dot or a cross. Backups are listed in one line, and manual steps get their own section.
  • The spinner is the tool's own, so Ctrl+C still cancels while an agent CLI runs.
  • Without a terminal, the output is plain text with the same lines.

agentbuffs 0.1.0

Choose a tag to compare

@AltanS AltanS released this 01 Nov 09:48

Added

  • First release. It adds the agentbuffs MCP server and the find-agent-tools skill to the agents on your machine, from the setup list at agentbuffs.com.
  • It edits JSON and TOML config files in place and keeps the rest of the file. It copies a file to <file>.bak first and writes through a temporary file.
  • It runs an agent's own CLI (for example claude mcp add) only from a fixed list, never through a shell.
  • It checks the skill file against its SHA-256 before it writes it.
  • The options are --client, --what, --yes, --dry-run, --api, --version and --help. The exit codes are 0, 1, 2 and 130.