# Dogfooding > Dogfooding workflow for improving Copilot customizations with evidence from the current repository. Use when creating or updating copilot-instructions.md, AGENTS.md, instructions, prompts, or skills. Use when this capability is needed. - Skill: `tomevault-io/dogfooding-2` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/dogfooding-2` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/dogfooding-2/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Productivity - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/dogfooding-2 --- # Dogfooding Customizations ## Outcome Produce a high-signal customization update grounded in real repository conventions, with minimal duplication and clear next actions. ## When to Use - Create or update workspace instructions. - Create prompts, skills, or agents from a workflow already used in practice. - Improve existing customization files that feel generic, stale, or duplicated. ## Procedure 1. Capture the target and scope. - Confirm what file should be produced or updated. - Decide scope: workspace (recommended for team use) or personal. - If the request is underspecified, ask only the minimum clarifying questions. 2. Discover existing conventions first. - Search for current customization files and core docs before writing anything. - Prefer linking to docs over repeating long content. - Detect whether a file should be created or merged. 3. Explore codebase facts with evidence. - Collect exact build, test, and lint commands. - Identify architecture boundaries and ownership by folder/file. - Capture project-specific gotchas that are not obvious from generic best practices. 4. Verify critical points from primary sources. - Validate subagent findings by reading key files directly. - Resolve inconsistencies before editing. 5. Generate or merge with minimal surface area. - Preserve useful existing sections. - Remove stale or duplicated guidance. - Keep instructions concise, actionable, and always-on only when truly global. 6. Add quality gates. - Ensure names/paths/frontmatter are valid for the customization type. - Ensure guidance is specific to this repo, not boilerplate. - Ensure references point to existing files. 7. Close the loop. - Summarize what was created or changed. - Suggest 3-5 realistic example prompts that exercise the customization. - Suggest related next customizations that naturally extend the workflow. ## Decision Points - Existing file found: merge and refresh; do not replace blindly. - No existing file: create the minimal template needed. - Repo has rich docs: link to docs; do not embed large excerpts. - Workflow unclear: ask for desired outcome, scope, and level of detail (checklist vs full workflow). ## Completion Checks - The produced file is in the correct location and naming format. - The guidance includes concrete repo commands and boundaries. - The result avoids duplication and follows link-first documentation. - The output includes clear examples for immediate use. --- > Source: [DevGeoRamos/my-bingo-mixer](https://github.com/DevGeoRamos/my-bingo-mixer) — distributed by [TomeVault](https://tomevault.io).