Skip to content

github-action.md

Latest commit

 

History

History
76 lines (58 loc) · 2.82 KB

github-action.md

File metadata and controls

76 lines (58 loc) · 2.82 KB
Raw

GitHub Action (planned)

Status: planned / design note. This describes a future surface. It is not yet released. The design reuses the existing SDK and CLI — no new analysis logic.

Goal

Bring read-only analysis into CI: when a workflow produces logs or changes manifests, analyze them and surface ranked root causes and read-only diagnostics right in the PR or job summary.

Proposed usage

name: devops-ai
on: [pull_request]

jobs:
  analyze:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Validate manifests
        uses: devopsaitoolkit/devops-ai-action@v1
        with:
          command: validate
          path: k8s/**/*.yaml

      - name: Analyze failing test logs
        if: failure()
        uses: devopsaitoolkit/devops-ai-action@v1
        with:
          command: analyze
          path: build/test.log
          tech: kubernetes

Proposed inputs

Input Description Default
command analyze | explain | validate analyze
path File or glob to read —
tech Technology hint auto-detect
enrich Add LLM narrative (requires a provider secret) false
provider anthropic | openai | gemini | ollama configured
fail-on none | match | invalid none

Proposed behaviour

  • Runs entirely offline by default — no secrets required for deterministic analysis.
  • Writes a Markdown report to the job summary ($GITHUB_STEP_SUMMARY) and, optionally, annotations on changed lines.
  • fail-on: invalid fails the job when validate finds errors; fail-on: match fails when a known failure signature is detected (handy for blocking known-bad patterns).
  • Enrichment is opt-in and reads a provider key from repository secrets, e.g. ANTHROPIC_API_KEY. See Configuration and Security.

Implementation sketch

A composite action that installs the package and shells out to the CLI with --json, then formats the result for the job summary:

pip install devops-ai-toolkit
devops-ai "$COMMAND" "$PATH" ${TECH:+--tech $TECH} --json > result.json
python format_summary.py result.json >> "$GITHUB_STEP_SUMMARY"

Because it's built on the shared engine, output matches the CLI, SDK, and API exactly.

Feedback

Want this sooner, or have requirements? Open an issue (see Contributing) or follow https://devopsaitoolkit.com/blog.