Portable, safety-aware workflows for AI coding agents. Capture a session before context disappears, audit a project without guessing, and turn a vague idea into a buildable plan.
Copy a prompt into any AI agent, install the bundled Codex plugin, or reuse a standalone skill in Claude Code.
| Workflow | Use it when | It produces |
|---|---|---|
| Session Handoff | You are ending a coding session | Durable project memory, decisions, validation results, and exact next steps |
| Deep Project Audit | You need to understand an unfamiliar or risky codebase | Architecture map, operational picture, prioritized findings, and a remediation plan |
| Idea Discovery Interview | You have only a rough product or feature idea | A recorded discovery brief and an evidence-based implementation plan |
| Security Red-Team Audit | You are authorized to security-test a project | Threat model, verified findings, safe remediation, and regression tests |
| Product Growth Audit | An existing app needs a clearer improvement direction | Evidence-based UI/UX, product, business, and technical roadmap |
Capture context → Understand the system → Build the right thing
git clone https://github.com/bglglzd/agent-workflows.gitOpen a file in prompts/, paste its Prompt section into your agent, and give it the current project directory. The prompts are plain Markdown by design: they work in chat, terminals, IDE agents, and project instructions.
This repository ships an installable Codex plugin containing all three skills at plugins/agent-workflows. Add its marketplace, then install the plugin:
codex plugin marketplace add bglglzd/agent-workflows --ref main
codex plugin add agent-workflows@agent-workflowsStart a new session, then invoke a workflow explicitly, for example: Use $session-handoff to preserve this project's context. Codex can also select a workflow implicitly when its description matches your task.
For a project-local skill without installing the plugin, copy an individual folder into .agents/skills/:
New-Item -ItemType Directory -Force .\.agents\skills | Out-Null
Copy-Item .\plugins\agent-workflows\skills\project-audit .\.agents\skills\project-audit -RecurseCopy a skill folder into the project-local .claude/skills/ directory. Keeping it in the repository makes the workflow available to collaborators too.
New-Item -ItemType Directory -Force .\.claude\skills | Out-Null
Copy-Item .\plugins\agent-workflows\skills\project-audit .\.claude\skills\project-audit -RecurseUse the canonical Markdown files from prompts/. Tools such as OpenCode or OpenClaw may use different command and instruction-file conventions; paste the relevant prompt into that tool's project instruction or custom command. This repository deliberately does not claim automatic installation where a tool has no compatible skill format.
- Evidence before conclusions. Prompts ask for file paths, commands, validation output, and explicit unknowns.
- Safe by default. Audits are read-only and never connect to servers or expose secrets unless the user explicitly authorizes that scope.
- Durable context. Handoffs preserve decisions and the exact state needed by the next human or agent.
- No fake certainty. If a project, server, or requirement cannot be verified, the workflow says so.
- Tool-neutral core. Markdown is the source format; skill wrappers add convenience, not lock-in.
New workflows are welcome when they solve a recurring engineering problem and include clear scope, safety boundaries, concrete outputs, and verification guidance. Read CONTRIBUTING.md before opening a pull request.
- Debugging and incident-response workflow
- Release-readiness workflow
- Migration and modernization workflow
- Community-tested examples for popular AI coding agents
MIT — see LICENSE.