# Workflow Work > Complete an authorized Emisar implementation with focused feedback, canonical final gates, and scope discipline. Use for a plan, checklist, or request to implement or continue work. - Skill: `andrewdryga/workflow-work` (Agent Skill) - Install (CLI): `npx skillmds@latest add andrewdryga/workflow-work` - Raw SKILL.md: https://api.skillmd.com/api/skills/andrewdryga/workflow-work/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Productivity - Author: AndrewDryga (https://skillmd.com/u/andrewdryga) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/andrewdryga/workflow-work --- # Complete the implementation Follow the root and touched project AGENTS.md. The user's request or accepted plan defines the complete deliverable. Use workflow-spec when design is needed, then continue under the existing authorization. Work in coherent steps. Give concise updates about findings and the next check. Use Elixir context/testing skills for Portal, go-engineer for runner/MCP, and security-engineer for trust boundaries. Verify unfamiliar APIs against source or documentation before using them. Run focused checks for the behavior being changed. For Portal, use focused Credo after coherent Elixir edits and relevant ExUnit tests. Keep required happy, denial, cross-account, and abuse cases. Fix a red check before dependent work; gather missing evidence and repair within scope instead of stopping at the first failure. Never weaken an assertion or gate to hide a defect. Adapt routine implementation details as evidence develops. Preserve the full agreed behavior and security contracts. For a material scope or authority change, complete independent work and ask for that decision. Nearby unrelated issues belong in a follow-up. Use targeted edits unless most of the file is changing. The lead reviews delegated changes and runs each touched project's local final gate on the final tree. Portal uses `./run gate portal --changed`; run the complete `./run gate portal` before a push or release. Other projects use `./run gate `. Delegates provide focused check evidence; a handoff does not itself require duplicate full gates. Rerun or broaden verification after changes, failures, or unresolved concerns. Manual, skill, and KB-only work needs `./run check agent-setup` plus `./run check docs`; executable tooling changes also need `./run gate tooling`. Finish the lifecycle required by AGENTS.md: focused commit, task notes, and completion state. Report actual outcomes and validation, distinguishing source completion from push, deployment, and live proof. Preserve pending work and user decisions across compaction. Do not end with a promise to perform an authorized step that remains undone.