Dev Workflow — Step 4: Implement
SkillAI & modelsImplement a feature from its PRD. Creates a work tree if needed, writes clean code following Google-level standards, and tests iteratively. Sub-skill of the /dev workflow.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the Dev Workflow — Step 4: Implement skill
What this skill tells your AI
The instructions your AI receives, as published by browseros-ai/browseros-agent in .claude/skills/dev4-implement/SKILL.md and read by ahel’s review.
You are implementing a feature from its PRD. Write code like a staff software engineer at Google — clean, simple, and well-structured.
Input
- Read
.llm/$ARGUMENTS/prd.mdfor the full PRD - Read
.llm/$ARGUMENTS/design.mdfor design decisions - Read
.llm/$ARGUMENTS/tmp_exploration.mdfor codebase context
Step 1: Set up work tree
Check if you are already in a git work tree (not the main tree):
git rev-parse --is-inside-work-tree && git worktree list
If you are in the main work tree (not a feature worktree), create one:
wt switch -c feat/$ARGUMENTS
If already in a feature worktree, continue in place.
Step 2: Plan implementation order
Break the PRD into small, testable implementation steps. Write the plan to .llm/$ARGUMENTS/tmp_impl_plan.md. Each step should be:
- Small enough to verify independently
- Ordered so that dependencies come first
- Testable (you can run something to verify it works)
Step 3: Implement step by step
For each step in the plan:
- Write the code
- Test it (run existing tests, or manually verify)
- Fix any failures before moving to the next step
Code Style Guide
Follow these rules strictly:
- No excessive console.log — Only log when it serves a clear purpose (errors, important state changes). Remove debug logs.
- Self-contained functions — Each function should do one thing. No function should exceed 20-30 lines.
- Logic grouping — Within a function, keep related lines of logic together without blank lines between them. Use a blank line only to separate distinct logical blocks.
- Comments — Only add a comment when the logic is not self-evident. The comment should explain why, not what. Additionally, sprinkle short one-line
//comments on roughly half the major logic blocks in a function — enough to skim the function and follow the flow without reading every line. Keep these brief (e.g.,// validate input,// build response payload). Not every block needs one, but the big chunks should be signposted. - Simple and direct — No premature abstractions. No over-engineering. Write the simplest code that solves the problem.
- Follow existing patterns — Match the conventions already in the codebase (naming, file structure, imports, error handling).
Step 4: Verify
After all steps are implemented:
- Run the full test suite
- Manually verify the feature works as described in the PRD
- Fix anything that fails — loop back to implementation until it passes
Step 5: Hand off
Tell the user implementation is complete, then immediately invoke /dev5-review $ARGUMENTS.
Signals
- GitHub stars
- 51
- Forks
- 40
- Last commit
- Jun 2026
Advanced
- Catalog kind
- skill
- Gateway key
dev4-implement- Source
- github.com/browseros-ai/browseros-agent