Install a workflow in a project and it brings the skills and steps it uses. Every step says what it can reach before you install it, and runs with no network of its own.
10 packages4 downloads1 publisher
Closes spam, labels duplicates of earlier issues, labels questions and feature requests, and for bug reports tries to reproduce the bug in a sandbox, then replies on the issue according to what it found.
For an issue triage reproduced, fixes the bug in a sandbox starting from the reproduction, opens a draft pull request from a branch of its own, and links it on the issue.
Keeps a snapshot of the default branch after the project's setup commands, so sandbox steps that run them start from it and fetch only their own commit. Hashes the files that decide it, and builds again when that hash, the setup commands or the image change, and before it expires.
Checks whether an issue duplicates an earlier one, by searching the repository's issues with gh in an empty sandbox: nothing is cloned.
Reproduces a reported issue: with a minimal failing test in the repository, or in an empty sandbox with a minimal app against the published package. Then verifies the reproduction from clean.
Fixes a reported bug starting from its reproduction, pushes the fix to a branch of its own and opens a draft pull request.
Asks a Workers AI decision model (Clef) one question about some text and branches on the answer.
Cuts a list or text into items and keeps the ones wanted, or part of each: the id after finding: in a pull request's labels, or every match of a pattern.
Looks up a GitHub user, such as a pull request's author, and what they are to the project's repository: its owner, a collaborator, a contributor or nobody, and their permission. For an If step that decides whose pull requests to review.
Hashes the files matching some patterns at a commit, read from GitHub: lockfiles, say. Its hash is a value for later steps, such as what a snapshot is keyed on.
For an issue triage reproduced, fixes the bug in a sandbox starting from the reproduction, opens a draft pull request from a branch of its own, and links it on the issue.
Cuts a list or text into items and keeps the ones wanted, or part of each: the id after finding: in a pull request's labels, or every match of a pattern.
Reproduces a reported issue: with a minimal failing test in the repository, or in an empty sandbox with a minimal app against the published package. Then verifies the reproduction from clean.
Asks a Workers AI decision model (Clef) one question about some text and branches on the answer.
Checks whether an issue duplicates an earlier one, by searching the repository's issues with gh in an empty sandbox: nothing is cloned.
Read its readme and files here. A workflow lists what it installs with it; a step, what it can reach through Alcora.
From the project’s Registry page. Alcora copies each package into the project and checks its sha-256, so a run never depends on the registry.
A workflow names a range for each package it uses, and runs the newest version installed that the range allows. Nothing updates by itself.
The registry takes no uploads. Put a package in a folder of a public repository, then open a pull request adding it to registry.json. A new version in its alcora.json becomes a release.
{
"name": "acme/release-notes",
"type": "workflow",
"repo": "acme/alcora-packages",
"path": "release-notes",
"ref": "main"
}