convez.ai
Skill
opendatahub-io

autofix-skills

opendatahub-io/autofix-skills

Ver no GitHub
BomPluginScore · de 0 a 100

Claude Code plugin providing orchestrator skills, prompt files, and state management scripts for the Jira autofix pipeline

Instalação

$npx convez-ai opendatahub-io/autofix-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
11
Acompanhando
0
Saúde
75%
Último commit
há 4 dias
Versão
—
Issues abertas
7
Contribuidores
6

README

A tradução para o português deste README ainda está na fila.

autofix-skills

Claude Code plugin for the Jira autofix pipeline. Provides orchestrator skills, agent prompt files, and deterministic Python scripts for automated bug fixing, CVE remediation, and ticket triage.

Overview

This plugin is the inner layer of the autofix pipeline, the skills that run inside the Claude Code container. The outer layer (Python orchestration, GitLab CI, ticket management) lives in jira-autofix and agentic-ci (generic CI framework).

Skills

SkillDescription
autofix-resolveOrchestrates end-to-end bug fixing: implement, review, evaluate loop (max 3 iterations)
autofix-cve-resolveCVE remediation across multiple repos with state-machine dispatch
autofix-triageAssesses bug tickets for AI autofix readiness (ready/needs_info/not_fixable)

Scripts

ScriptDescription
merge_findings.pyMerges core review findings with extension findings, tags each with source
state.pyState persistence utility for context-compression recovery
cve_pipeline.pyCVE state machine: deterministic routing, wave dispatch, progress polling

Installation

This plugin is pre-installed in the ghcr.io/opendatahub-io/ai-helpers container image. For local development:

git clone git@github.com:opendatahub-io/autofix-skills.git

# Install as a Claude Code plugin (local directory source)
claude plugin install /path/to/autofix-skills

Development

Prerequisites

  • Python 3.10+ with ruff
  • shellcheck
  • Container runtime (Podman or Docker) for claudelint

Validate Changes

make lint

See CONTRIBUTING.md for the full development workflow.

Evaluation

The eval/ directory contains test cases for evaluating skill quality using the agent-eval-harness. Two eval configs are provided:

ConfigSkillCases
eval.yamlautofix-resolveeval/cases/ (10 cases: bug fixes, iteration, triage decisions, security guardrails)
eval-triage.yamlautofix-triageeval/cases-triage/ (4 cases: ready, needs_info, not_fixable, borderline)

Running evals

# Install the eval harness
pip install -e ../agent-eval-harness

# Run the autofix-resolve eval suite
/eval-run --config eval.yaml --model opus

# Run the triage eval suite
/eval-run --config eval-triage.yaml --model opus

# Run a single case
/eval-run --config eval.yaml --model opus --cases case-006-agents-md-compliance

Results are written to eval/runs/<run-id>/ with per-case verdicts, modified files, events, and an HTML report. See eval.md and eval-triage.md for detailed analysis docs.

Test case structure

Each case directory contains an input.yaml, .autofix-context/ with ticket data, optional source files in src/, and annotations.yaml with expected outcomes. AGENTS.md files are stored with a .fixture suffix to prevent agents working in this repo from auto-discovering test payloads (case-010 contains a credential-harvesting guardrail test). The eval SessionStart hook activates fixtures before each case runs.

Architecture

See AGENTS.md for architecture details and conventions.

Team-specific extension skills go in ai-helpers, not this repo. Extensions are normal Claude skills that read from .autofix-context/ and write findings to .autofix-context/extension-findings/<skill-name>.json.

Versioning

Use git tags (v0.1.0, v0.2.0, etc.) for releases. The outer-layer runner pins a release via SkillConfig.skill_ref. The main branch is the development head.

License

Apache License 2.0

Ver o original no GitHub

Bom para

Relacionados