auto-issue

Resolves GitHub issues end-to-end by writing a failing acceptance check, fixing the code, and opening a pull request.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve? Turning a GitHub issue into a reviewed, verified pull request normally requires manual triage, reproduction, fixing, testing, and PR authoring. This Skill automates that entire loop inside a sandboxed agent environment, while enforcing trust checks on issue authors and proof that the fix actually works. ## Core Features & Use Cases - Author trust verification: Checks the issue author's authorAssociation before acting, and stops for untrusted authors unless explicitly told to continue. - Triage before coding: Distinguishes issues needing a code change from ones that are already fixed, need more information, or should be closed as not planned. - Acceptance-check-first workflow: Writes a runnable check that fails before the fix and must pass after, so the PR carries evidence rather than a plausible-looking diff. - Sandboxed clone, gate, and PR: Shallow-clones the repository with gh credential helper auth, runs the repo's own lint/test gate via gate.sh, enforces a ~500-line diff size gate, and opens the PR with Closes #N. - Use Case: A maintainer says "fix issue 42". The Skill verifies the author, reproduces the bug with a failing test, implements the minimal fix, confirms the test flips to green, runs the repo's own checks, and opens a pull request linked back to the issue. ## Quick Start Ask the agent to take issue 42 to a pull request, for example by saying "/auto-issue 42" or "fix issue 42" on a task bound to the target repository.

Frequently Asked Questions about auto-issue

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

FAQPage Schema
How do I automatically fix a GitHub issue and open a pull request?▼

Invoke the skill with a phrase like "/auto-issue 42" or "fix issue 42" on a task bound to the repository. It triages the issue, writes a failing acceptance check, implements the minimal fix, verifies the check passes, and opens a PR with Closes #42 in the body.

How does the skill verify a GitHub issue author is trusted?▼

It reads the issue's authorAssociation field via gh issue view. OWNER, MEMBER, and COLLABORATOR proceed directly; CONTRIBUTOR, NONE, or FIRST_TIME_CONTRIBUTOR cause the run to stop and ask before continuing, since the token lacks permission to query collaborator endpoints.

What sandbox environments does this skill require?▼

It requires the E2B code sandbox, which provides the gh CLI and a repository-scoped token. It cannot run on a Sprites sandbox because gh is absent there. Commands execute through capsule.proc.exec with mandatory timeouts.

Why does gate.sh exit with code 2 or 3?▼

Exit code 2 means no lint, typecheck, or test command 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 run must stop until that host is allowed on the agent.

What happens when the fix diff is too large?▼

The skill enforces a size gate of roughly 500 changed lines measured against origin/HEAD. Above that threshold it stops and asks, suggesting the work be split into stacked PRs or narrowed in scope, because review quality degrades on large diffs.

When should I not use this skill on a GitHub issue?▼

Do not use it for reading or summarizing issues, only for resolving them. It also stops when triage is ambiguous, when no acceptance check can be made to fail, when a force-push would be needed, or when the issue does not actually require a code change.