Ecosystem Integrations — Hook‑Agents that play nice with your IDE & CI

Steps

1) Enable pre-commit hooks as above.
2) Add a small script to emit hook results to JSON.
3) In Claude Code, set a custom command to read the JSON and propose a patch.
4) Accept patch, re-run commit.

Emit hook results to JSON

# package.json
{
  "scripts": {
    "verify:json": "npx autodevops verify --staged --format json > .autodevops/out.json"
  }
}

# .husky/pre-commit
npm run verify:json || true

How it works

• Hook runs, writes .autodevops/out.json
• Claude reads JSON and crafts a targeted diff
• Developer approves; commit proceeds

Steps

1) Produce a unified diff of staged files.
2) Feed diff + error summary to your model/tooling.
3) Apply suggested patch and re-stage.
4) Commit.

Create a diff + prompt locally

# scripts/mkdiff.sh
#!/usr/bin/env bash
git diff --staged > .autodevops/diff.patch

# scripts/prompt.txt (consumed by your CLI of choice)
Context: This is a pre-commit failure summary and the staged diff.
Goal: produce the smallest possible patch to pass hooks.

How it works

• Hook provides failure summary
• Model proposes minimal patch
• Developer applies and commits

Steps

1) Add a post-failure runner in your pre-commit.
2) Pipe .autodevops/out.json + diff to Gemini CLI.
3) Review patch, apply if safe.

Pre-commit with Gemini handoff

# .husky/pre-commit (excerpt)
npx autodevops verify --staged --format json > .autodevops/out.json || FAIL=1
if [ "$FAIL" = "1" ]; then
  # pseudo: gemini CLI invocation
  echo "Handing off to Gemini..."
  # gemini --model=2.5-pro --prompt-file=scripts/prompt.txt --input=.autodevops/out.json .autodevops/diff.patch
  exit 1
fi

How it works

• On failure, shell hands context to Gemini
• Gemini suggests patch or rationale
• You apply and re-run commit

Steps

1) Ensure .autodevops/out.json is produced on each commit attempt.
2) Teach Cursor to watch that file (Custom Commands).
3) Map each issue to a fix-it instruction.

Cursor command idea

# .cursor/commands.json (conceptual)
[{
  "name": "Explain Hook Failures",
  "command": "read .autodevops/out.json and propose smallest patch for each item"
}]

How it works

• Background agent reacts to JSON updates
• Shows explanation + patch
• You accept and re-stage

Steps

1) Add a VS Code task that runs verify:json.
2) Expose the JSON path to Cline as context.
3) Use Cline's 'apply patch' to modify files.

VS Code task example

// .vscode/tasks.json
{
  "version": "2.0.0",
  "tasks": [{
    "label": "AutoDevOps Verify JSON",
    "type": "shell",
    "command": "npx autodevops verify --staged --format json > .autodevops/out.json"
  }]
}

How it works

• Task populates the JSON context
• Cline reads and proposes steps
• Tool runs apply the edits

Steps

1) Emit JSON + unified diff.
2) Let Augmentum read both as inputs.
3) Constrain it to 'smallest patch wins'.

Telemetry file shape

// .autodevops/out.json (shape)
{
  "issues": [
    { "agent": "secret-guard", "severity": "high", "path": ".env", "message": "API key detected" },
    { "agent": "test-focus-blocker", "severity": "med", "path": "auth.test.ts", "message": "it.only found" }
  ]
}

How it works

• Agent ingests issues + diff
• Proposes minimal patch set
• You review and commit

Steps

1) Add a CI workflow that runs 'autodevops verify' on PR diffs.
2) Fail PR if violations remain.
3) Post annotations as PR comments.

GitLab CI example

# .gitlab-ci.yml
verify:
  image: node:20
  stage: test
  script:
    - npm ci
    - npx autodevops verify --diff $CI_MERGE_REQUEST_DIFF_BASE_SHA..$CI_COMMIT_SHA

How it works

• Same checks in hooks and CI
• CI is the source of truth
• Green CI == mergeable