A repo-local Codex plugin marketplace for reusable agent workflows. The marketplace currently packages Agent Ops investigation and documentation helpers, Agent Dev project-shaping and cleanup workflows, Amazon account workflows, Subtractive UI frontend-review workflows, Agent SDLC development workflows, and Rehearse communication practice as installable Codex plugins.
This repo exposes its Codex marketplace at:
.agents/plugins/marketplace.json
From a local checkout, add the marketplace source to Codex:
codex plugin marketplace add /path/to/agent-pluginsAfter changes are pushed, add a Git marketplace and choose its ref:
codex plugin marketplace add jlave-dev/agent-plugins --ref mainRefresh a configured Git marketplace, then install a plugin from its snapshot:
codex plugin marketplace upgrade <marketplace-name>
codex plugin add <plugin-name>@<marketplace-name>Operational helpers for evidence-led investigation, repo documentation, note and plan capture, and prompt drafting.
Included Codex skills:
reason-from-evidence: test hypotheses against facts, observations, reports, assumptions, and constraints without overstating conclusions.create-agents-md: create or refresh high-signalAGENTS.mdfiles.create-readme: create or refresh useful projectREADME.mdfiles.save-note: save relevant thread context as local Markdown notes.save-plan: save durable planning documents as local Markdown files.write-prompt: turn rough asks into copy-ready prompts, from a short paragraph for simple tasks to selective structure for complex work.
Developer workflows for shaping new software projects and simplifying existing repos.
Included Codex skills:
map-project-decisions: map the first usable slice and separate human-owned product or architecture gates from safe agent decisions, empirical questions, and deferred work.yagni: audit or remove stale, unused, duplicated, speculative, or misleading repo surfaces.
Amazon workflows for product research, order inspection, and review drafting.
Included Codex skills:
shop-amazon: verify Amazon product-page facts and return ranked product recommendations.find-orders: inspect Amazon orders, invoices, return windows, warranty clues, and reorder availability.write-reviews: draft Amazon product reviews and submit only after explicit approval.
Minimal, reference-led frontend critique for auditing and refactoring product UI.
Included Codex skills:
audit-rendered-ui: review real rendered UI for leaky copy, redundancy, layout issues, and handoff polish.refactor-ui: remove UI clutter and implementation-detail copy while preserving behavior.reference-ruler-ui: compare an implementation against a concrete visual reference with precise layout and styling checks.
Zero-config issue intake, worker dispatch, evidence refresh, review, and merge coordination for mostly agentic software development.
Included Codex skills:
sdlc-orchestrate: inspect issue/PR state and continue to the next SDLC role automatically.sdlc-issue-intake: turn rough requests into scoped GitHub Issues with saved worker dispatch prompts and declared CI tiers.sdlc-preflight: verify branch, worktree, base, and active-work overlap before dispatch.sdlc-dispatch-issue: pick a ready SDLC issue, create its branch/worktree, launch the worker thread, and record the worker state on the issue.sdlc-evidence: refresh PR evidence, current-head checks, attached proof, and issue Agent State.sdlc-review-loop: send current changes through an implementer/reviewer loop with CI evidence expectations.sdlc-merge-queue: merge approved SDLC PRs in dependency order without fixing worker branches inline.sdlc-reviewer: review a change handoff and return an explicit verdict.
Practice explaining real work to a maintainer or stakeholder. The rehearse skill researches the relevant context, asks one question at a time, follows gaps, and produces questions to revisit, facts to study, and a concise answer bank in your own words. Supports PR, status, design, incident, and executive rehearsals, with thinking time and focused retries.
Example: Use $rehearse to help me explain this branch to its maintainer.
The plugin launches the pinned VoiceMode MCP server automatically through uvx and uses the computer's microphone and speakers directly. No browser Playground or manually started MCP server is needed. One-time setup uses VoiceMode's upstream installer:
uvx voice-mode-install --yes
uvx --from voice-mode==9.0.2 voicemode service install whisper
uvx --from voice-mode==9.0.2 voicemode service install kokoroThe installer provisions system dependencies; local service installation downloads models and may register background services. Then reconnect the host's MCP connection and allow microphone access when prompted. Rehearse configures local speech endpoints with audio/transcript saving disabled. See the VoiceMode setup and conversation reference for service checks, startup, other harnesses, and troubleshooting. Active native voice or text can also be used. The skill is portable Markdown; the marketplace packaging targets Codex.
The repository has no declared runtime or development dependencies. Run the test suite directly:
npm testValidate a skill after editing its SKILL.md, references, or agents/openai.yaml:
python3 <codex-home>/skills/.system/skill-creator/scripts/quick_validate.py plugins/<plugin-name>/skills/<skill-name>Validate a plugin after editing its manifest, marketplace metadata, or skill layout:
python3 <codex-home>/skills/.system/plugin-creator/scripts/validate_plugin.py plugins/<plugin-name>Use a cachebuster only when testing an installed plugin through local Codex. It forces Codex to load changed skills, scripts, or assets under a new cache key:
python3 <codex-home>/skills/.system/plugin-creator/scripts/update_plugin_cachebuster.py plugins/<plugin-name>Revert the manifest version before committing. Include a cachebuster in a PR only when the PR tests cache behavior or reviewers must install that exact branch.
agent-plugins/
├── .agents/
│ └── plugins/
│ └── marketplace.json # Codex plugin marketplace
├── plugins/
│ ├── amazon/
│ │ ├── .codex-plugin/
│ │ │ └── plugin.json
│ │ └── skills/
│ ├── agent-dev/
│ │ ├── .codex-plugin/
│ │ │ └── plugin.json
│ │ ├── assets/
│ │ └── skills/
│ ├── agent-ops/
│ │ ├── .codex-plugin/
│ │ │ └── plugin.json
│ │ ├── assets/
│ │ └── skills/
│ ├── agent-sdlc/
│ │ ├── .codex-plugin/
│ │ │ └── plugin.json
│ │ ├── assets/
│ │ ├── scripts/
│ │ └── skills/
│ ├── rehearse/
│ │ ├── .codex-plugin/
│ │ │ └── plugin.json
│ │ ├── .mcp.json
│ │ ├── assets/
│ │ └── skills/
│ └── subtractive-ui/
│ ├── .codex-plugin/
│ │ └── plugin.json
│ └── skills/
├── scripts/ # Release helper scripts and tests
├── .github/workflows/ # Release automation
├── .releaserc.json # semantic-release configuration
├── dist/ # Ignored release artifacts
└── package.json # Development package metadata
This project uses automated semantic versioning powered by semantic-release. Every merge to main is evaluated automatically and, when commits require it, semantic-release creates:
- a Git tag (
vX.Y.Z) - a GitHub Release with generated notes
- a downloadable archive named
agent-plugins-vX.Y.Z.tar.gz
Release notes live in the repository's GitHub Releases page; there is no committed CHANGELOG.md. During release preparation, semantic-release rewrites package.json and plugin manifests to the computed release version, packages the archive, and commits those version bumps back to main with [skip ci].
This project follows Conventional Commits to drive automated releases:
<type>[optional scope]: <description>
Common types:
feat: new skill, plugin, or featurefix: bug fixdocs: documentation changeschore: maintenance taskstest: test additions or changes
Use the plugin name for plugin-wide changes and the skill name for skill-specific changes:
feat(amazon): add order workflow plugin
fix(shop-amazon): handle empty search results
docs(write-reviews): clarify submit approvalDo not include codex in branch names or commit messages.
MIT. See TERMS.md and PRIVACY.md for repository-level usage and privacy notes.