Autonomous Coding Workflow

SkillDev tools

Lets your agent make code changes across a repository in small verified steps with tests and checks after each edit.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Autonomous Coding Workflow skill

About this capability

Governs safe, incremental, repository-wide development workflow with continuous validation gates

What this skill tells your AI

The instructions your AI receives, as published by bottelet/daybydaycrm in .claude/skills/autonomous-coding-workflow/SKILL.md and read by ahel’s review.

Goal

Perform repository-wide modifications safely, incrementally, and with continuous validation.


1. Instruction Precedence

Before doing anything:

  • Check for repository-level instruction files:
    • .github/copilot-instructions.md
    • AGENTS.md
    • .junie/*.md
    • CLAUDE.md

If they exist:

  • Treat them as higher precedence for architecture and conventions.
  • Avoid duplicating rules already defined there.

2. Preparation

Before modifying code:

  1. Read existing implementation.
  2. Understand current behavior.
  3. Identify existing abstractions and reuse them:
    • Traits
    • Base test cases
    • Base resources
    • Base seeders
    • Services
    • DTOs
    • Transformers
  4. Search explicitly for duplication before introducing new abstractions.
  5. Preserve existing architectural patterns.

Do not modify code that has not been understood.


3. Refactoring Heuristics

Apply only when relevant:

  • If repeated patterns exist across many test classes, models, or resources, evaluate abstraction opportunities.
  • Prefer centralizing duplicated logic into:
    • Traits
    • Base classes
    • Services
  • Do not introduce abstraction unless duplication is confirmed.

4. Incremental Development

Work in small, verifiable steps.

After each change:

  1. Verify syntax:
    php -l
    
  2. Run targeted tests.
  3. Fix failures immediately.
  4. Run code style checks:
    vendor/bin/pint --dirty --format agent
    
  5. Continue only if repository is clean.

5. Validation Gates

Never proceed if any of the following fail:

  • PHPUnit tests
  • Static analysis
  • PHP syntax check (php -l)
  • Code style (Pint)

Before completion, additionally ensure:

  • migrate:fresh --seed passes
  • smoke tests pass
  • targeted tests pass
  • full suite passes (unless explicitly excluded)

6. Module Completion

After completing a module:

  1. Run targeted test suite.
  2. Run php -l.
  3. Run Pint.
  4. Confirm no unintended changes.
  5. Commit with clear module description.

Do not start the next module until the current one is fully stable.


7. Uncertainty Handling

If behavior is unclear:

  • Stop immediately.
  • Describe ambiguity.
  • Request clarification.
  • Do not infer or guess missing business rules.

8. Success Criteria

The task is complete only when:

  • Behavior is preserved.
  • No duplicate logic introduced.
  • Changes are idempotent.
  • All tests pass.
  • Full suite passes.
  • Formatting is clean.
  • No unintended architectural drift occurred.

Signals

GitHub stars
2k
Forks
767
Last commit
Aug 2026
Advanced
Catalog kind
skill
Gateway key
autonomous-coding-workflow
Source
github.com/bottelet/daybydaycrm