# Develop > Load before authorized code changes: features, fixes, refactors, tests, and mechanical edits, including discussion-to-coding transitions. Read separately from edits. Skip read-only requests and implementation already owned by a specialist. - Skill: `flurdy/develop` (Agent Skill) - Install (CLI): `npx skillmds@latest add flurdy/develop` - Raw SKILL.md: https://api.skillmd.com/api/skills/flurdy/develop/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: flurdy (https://skillmd.com/u/flurdy) - Updated: 2026-09-10 - Page: https://skillmd.com/skills/flurdy/develop --- # Develop A lightweight coding entry point, not another implementation workflow. Request the runtime's configured standard/high route, then use normal repository-grounded coding judgment. Explicit invocation is `/develop `; in Pi use `/skill:develop `. On Claude Code this skill carries no `model:` pin on purpose: it rides the session model rather than downgrading to the standard alias, while Pi routes by `model-tier`. Read this skill once per new coding run, even if its text remains in context from a previous run. Use a standalone skill read before generating mutation calls. Wait for the next model response before emitting edits, writes, or shell mutations; do not bundle the read and mutations in the same response. Apply the same separation when loading a specialist below. Already-generated edits are not upgraded by a model switch. ## Choose the next step - Keep read-only requests read-only. Skill selection does not authorize edits or override plan mode, repository rules, scope, or confirmation requirements. - If an implementation specialist already owns the work, continue it; do not restart through this entry point just to request a route. - If a bug's cause is unknown, use [diagnose-bug](../diagnose-bug/SKILL.md) first. - If consequential architecture, security, public-contract, migration, or hard-to-reverse decisions remain unresolved, use [architect](../architect/SKILL.md) before coding. - For implementation-ready work with interacting behavior, meaningful state/error paths, or local design trade-offs, load [implement-solution](../implement-solution/SKILL.md). Reassess if an initially simple edit develops those characteristics; hand off without restarting completed work. - Otherwise, make the smallest appropriate change using established repository patterns and proportionate tests/checks. Do not add planning, delegation, or review ceremonies merely because this skill loaded. ## Routing limits Automatic loading is best-effort, not a pre-write enforcement mechanism. Metadata requests capability; it does not prove the active model changed. Respect runtime spend and manual-selection controls; do not edit routing configuration or launch another model to force a route. Pi currently protects against nested tier downgrades only after a routed run begins. The first standard-tier route can replace a stronger session model. This skill does not establish a capability floor; see [model routing](../../MODEL_ROUTING.md).