code-execution

Code Execution

Overview

Execute coding subtasks with self-review and quality validation. Runs in isolated coder-agent context with pre-loaded standards.

Announce at start: "I'm using the code-execution skill to implement [subtask title]."

The Process

Step 1: Read Subtask JSON

Load the subtask file specified in $ARGUMENTS:

Read: .tmp/tasks/{feature}/subtask_{seq}.json

Extract:

• title — What to implement
• acceptance_criteria — What defines success
• deliverables — Files/endpoints to create
• context_files — Standards to apply
• reference_files — Existing code to study

Step 2: Load Context Files

Read each file in context_files:

Read: .opencode/context/core/standards/code-quality.md
Read: .opencode/context/core/standards/security-patterns.md

Understand:

• Project coding standards
• Naming conventions
• Security patterns
• Code quality requirements

Step 3: Load Reference Files

Read each file in reference_files:

Read: src/middleware/auth.middleware.ts

Study:

• Existing patterns
• Code structure
• Conventions in use

Step 4: Update Status to In Progress

Edit subtask JSON:

"status": "in_progress",
"agent_id": "coder-agent",
"started_at": "2026-02-16T00:00:00Z"

Step 5: Implement Deliverables

For each deliverable:

• Create or modify the specified file
• Follow acceptance criteria EXACTLY
• Apply standards from context_files
• Use patterns from reference_files
• Write clean, modular, functional code

Step 6: Run Self-Review (MANDATORY)

Type & Import Validation:

• ✅ Function signatures match usage
• ✅ All imports/exports exist
• ✅ No missing type annotations
• ✅ No circular dependencies

Anti-Pattern Scan:

grep "console.log" deliverables  # NO debug statements
grep "TODO\|FIXME" deliverables  # NO unfinished work
grep "api[_-]key\|secret" deliverables  # NO hardcoded secrets

Acceptance Criteria Check:

• Re-read acceptance_criteria array
• Confirm EACH criterion met
• Fix unmet criteria before proceeding