Skip to content

cloud-agent-usage.md

Latest commit

 

History

History
102 lines (69 loc) · 2.1 KB

cloud-agent-usage.md

File metadata and controls

102 lines (69 loc) · 2.1 KB
Raw

Cloud agent usage

Use /ppp-cloud for autonomous coding agents.

Prerequisites: The cloud agent runs on remote infrastructure and cannot access personal skill installs. Run npx ai-engineering-starter-kit install --repo-local in the target repo first to commit the skill into .github/skills/. See Repo-local install.

The slash form /ppp-cloud is the user-facing invocation. The installed skill directory and identifier remain ppp-cloud.

It is the non-interactive version of Plan. Patch. Prove.

Inspect → Decide → Plan → Prove → Patch → Validate → Review → Draft PR

Good cloud-agent tasks

  • clear bug fixes
  • test additions
  • small validation rules
  • small UI states
  • docs updates
  • small refactors with tests
  • coverage around existing behaviour

Avoid assigning directly

  • broad features
  • architecture-heavy work
  • auth/security/permission changes
  • database migrations
  • public API decisions
  • billing/tenancy changes
  • unclear product behaviour

Expected output

The cloud agent should either:

  1. create one focused draft PR; or
  2. stop with a useful blocker.

Suggested issue/task format

Use the `/ppp-cloud` skill.

## Task

<clear bounded task>

## Expected behaviour

- ...

## Scope

Do:
- ...

Do not:
- ...

## Context

- Relevant files/modules:
- Existing pattern to follow:
- Related tests:

## Proof required

- Add/update:
- Run:

## Stop conditions

Stop and report if:
- product behaviour is unclear;
- auth/security/data/API changes are required;
- tests fail after two focused attempts;
- implementation requires broad refactoring;
- repo instructions conflict with this task.

## PR expectation

Create one focused draft PR.
Follow repo PR template and commit/title conventions.

This repo also includes a ready-to-copy issue template at .github/ISSUE_TEMPLATE/ppp-cloud-task.md.

Review expectation

Cloud-created PRs should still be reviewed by a human.

Focus review on:

  • assumptions
  • tests/checks run
  • checks not run
  • scope creep
  • security/privacy/auth/data/API risks
  • whether the PR should remain draft