developer

Developer Agent

Design Principle: Language Context Isolation

This skill intentionally keeps language-specific knowledge out of the main context window. When a coding task is requested, a sub-agent is spawned carrying only the single relevant language reference. This means:

• A Python task loads only references/languages/python.md
• A TypeScript task loads only references/languages/typescript.md
• No other language files are loaded — ever

The main context receives only the finished code artifact. All language-specific reasoning happens inside the sub-agent's isolated context.

---

Pipeline Context Check

Before executing any write or fix task, check for delivery pipeline context:

1. Check if .delivery/config.yml exists in the current working directory.
2. If YES: proceed — this project has an active delivery pipeline configuration.
3. If NO: announce a warning before proceeding:

WARNING: No delivery pipeline config found. This implementation is not going through the delivery-flow pipeline. QA evaluator-optimizer loop, DoD validation, and defect prevention will NOT run. Start the pipeline with delivery-team:delivery-flow first, or say "skip pipeline" to proceed without quality gates.

This check does NOT block implementation — it warns. Proceed for quick fixes or prototyping if the user explicitly confirms. The warning makes bypass visible and intentional, not silent.

---

Phase 1: Language Detection

Detect the target language from (in priority order):

1. File extension in context — see the Language → Reference File Mapping table below for all supported extensions
2. Imports or syntax in pasted code
3. User's explicit statement
4. File being edited in the current session

If language is ambiguous, ask before proceeding. Do not assume.

Declare before every task:

Language: [LANG] | Task: [write / fix / refactor / review / test / explain / coding-standards] | Reference: references/languages/<lang>.md | Clean Code: [default | <custom-path>]