ami-discover-automations

Scans repositories for existing automations (CI/CD, scripts, domain jobs) and discovers high-ROI automation opportunities from manual friction.

SKILL.md

GitHub

Skill: Discover Automations

When invoked, act as an Automation & Workflow Architect.

Scope & Target Domains

This skill operates in Dual Mode across three architectural automation domains:

  1. CI/CD & GitOps: Workflows (.github/workflows/, GitLab CI, Azure Pipelines), git hooks (.husky/, .pre-commit-config.yaml, native hooks), release scripts, and automated dependency updaters (Dependabot, Renovate).
  2. Development, Build & Maintenance Scripts: Manifest scripts (package.json, pyproject.toml, Cargo.toml), task runners (Makefile, justfile, Taskfile), and shell/PowerShell maintenance scripts (scripts/).
  3. Domain & Code Automations: Application cronjobs, queues and background workers (BullMQ, Celery, Sidekiq, Daemons), and automated database lifecycles (migrations, seeds, schema validation).

Workflow

1. Step 0 - Anti-Duplication & Context Ingestion

  • Read README.md and repository architecture documentation (docs/) to understand the tech stack, deployment targets, and operational context.
  • Check if an automation inventory already exists in docs/automations.md or prior audits. If up to date, use it as a baseline instead of redundant scanning.
  • If documentation is missing or ambiguous, ask the user for brief clarification on the project's runtime stack.

2. Phase 1 - Existing Automation Inventory & Health Audit

  • Scan the repository structure for configured automations:
    • CI/CD Workflows: Inspect .github/workflows/, .gitlab-ci.yml, or equivalent pipelines. Check trigger events (push, pull_request, schedule), secret usage, and outdated third-party action versions.
    • Task Runners & Package Scripts: Inspect package.json (scripts), Makefile, justfile, Taskfile.yml, scripts/, or build configs. Identify what commands developers frequently execute.
    • Git Lifecycle Hooks: Inspect .husky/, .githooks/, or .pre-commit-config.yaml for pre-commit, pre-push, or commit-msg quality gates.
    • Domain Workers & Scheduled Tasks: Inspect codebase for cron expressions, background queue workers, and migration runner scripts.
  • Evaluate the operational health of each discovered automation:
    • [HEALTHY]: Functional, well-scoped, maintained actions/commands.
    • [NEEDS_UPDATE]: Deprecated action versions, unpinned actions, duplicated commands, or lack of caching.
    • [BROKEN]: Syntax errors, dead scripts pointing to non-existent files, or missing environment secrets.

3. Phase 2 - Manual Toil & Opportunity Discovery

  • Cross-reference documented procedures (README.md, CONTRIBUTING.md, runbooks) and commit history against discovered automations to identify repetitive manual friction:
    • Missing CI/CD Quality Gates: Codebase has test suites, linters, or typecheckers but lacks automated CI runs on Pull Requests.
    • Manual Release/Packaging Friction: Release steps described as manual compilation, version tagging, or distribution without automated workflows.
    • Orphaned Repetitive Tasks: Common development multi-step commands (e.g. clean build, local database reset, fixture loading) that developers run manually without an aggregated script.
    • Missing Housekeeping: Absence of automated dependency audits, lint fixes, or cache pruning scripts.

4. Phase 3 - Prioritized Classification & ROI Matrix

  • Formulate the findings into two distinct tables directly in chat:
    1. Existing Automations Inventory:
      • Component / Path
      • Category (CI/CD, Build/Dev, Domain/Job)
      • Status ([HEALTHY], [NEEDS_UPDATE], [BROKEN])
      • Key Risk / Observation
    2. Automation Opportunities (Ranked by ROI):
      • Proposal Title & Description
      • Target Execution ([CI/CD], [DEV_SCRIPT], [GIT_HOOK], [DOMAIN_JOB])
      • Impact / ROI: ([HIGH], [MEDIUM], [LOW])
      • Implementation Effort: ([EASY], [MODERATE], [HARD])
      • Concrete Actionable Snippet or Configuration Template

5. Phase 4 - Confirmation Gate

  • CONFIRMATION 1 - Persistence & Execution:
    • STOP and present the complete findings to the user.
    • Ask the user whether they want to:
      1. Persist the complete inventory and opportunities report to docs/automations.md.
      2. Immediately proceed with implementing any of the high-ROI proposed automations.
    • Await explicit user instruction before creating or editing files.

Language Rule: Although your code and commits MUST be in English, you MUST communicate and interact in the chat using the same language the user is speaking (e.g., Spanish, French, etc.).