auto-fix

Reproduce reported bugs, fix them minimally, and open verified pull requests on GitHub.

Updated Mar 20, 2026
One-click install
npx skills add https://github.com/astralform-ai/skills --skill auto-fix-astralform-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: auto-fix
Source: https://github.com/astralform-ai/skills/tree/main/skills/auto-fix
Command: npx skills add https://github.com/astralform-ai/skills --skill auto-fix-astralform-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Turning a vague bug report into a working fix is slow and error-prone: symptoms get misdiagnosed, patches ship without proof, and reviewers cannot tell whether a change actually does anything. This Skill takes a reported symptom all the way to a pull request inside a sandboxed environment, with evidence at every step. ## Core Features & Use Cases - Reproduction-first workflow: Clones the repository, reproduces the symptom, and converts it into a failing acceptance check before any code is changed. - Minimal, verified fixes: Finds the root cause, fixes only what the symptom requires, then proves the acceptance check flips from failing to passing. - Repository gate execution: Runs the repo's own lint, typecheck, and test commands via gate.sh, distinguishing real failures from sandbox egress limits or missing toolchains. - Size-gated PR creation: Stages only intended changes, enforces a roughly 500-line review limit, and opens a pull request against the repository's default branch. - Use Case: A user reports "the export button does nothing." The Skill clones the repo, writes a failing test for the export path, fixes the underlying null handling, watches the test pass, runs the repo's test suite, and opens a PR documenting the symptom, cause, and verification. ## Quick Start Ask the agent to auto-fix the symptom you are seeing, for example: "/auto-fix the sidebar flickers on reload in this repository."

Frequently Asked Questions about auto-fix

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I turn a bug report into a pull request automatically?

Describe the symptom with the exact input, observed behavior, and expected behavior, then invoke the auto-fix workflow. It clones the repository, reproduces the symptom as a failing check, fixes the cause, verifies the check passes, and opens a pull request.

How does the skill verify a bug fix actually works?

It converts the reproduced symptom into an acceptance check and watches that check fail before the fix and pass after it, in the same run. It then runs the repository's own lint, typecheck, and test scripts via gate.sh as an additional gate.

What environments does this bug-fixing workflow require?

It requires a sandbox with the gh CLI installed and a GH_TOKEN scoped to the target repository, such as the E2B code sandbox. A Sprites sandbox lacks gh and is not supported, and network egress is limited to github.com plus allowed hosts.

Why does gate.sh exit with code 2 or 3 instead of passing?

Exit code 2 means no lint, typecheck, or test could run, either because the repo defines none or the sandbox lacks the toolchain. Exit code 3 means a dependency install hit a blocked host, and the output names the host that must be allowed.

When will the auto-fix workflow stop without opening a PR?

It stops when the symptom cannot be reproduced, the acceptance check will not fail beforehand, the cause lies in a dependency or another repository, the diff exceeds roughly 500 lines, tests cannot run due to egress limits, or a force-push would be required.