Central marketplace for Claude Code and OpenAI Codex CLI plugins by xiaolai.
Two manifest files live in this repo:
.claude-plugin/marketplace.json— Claude Code entries (read byclaude plugin install).agents/plugins/marketplace.json— Codex CLI entries (read bycodex plugin install)
Each plugin repo can ship both layouts in parallel: .claude-plugin/ for Claude Code, .codex-plugin/ + codex/ tree for Codex. The two ecosystems coexist in one source repo per plugin.
For Claude Code:
claude plugin marketplace add xiaolai/claude-plugin-marketplaceFor Codex CLI:
codex plugin marketplace add xiaolai/claude-plugin-marketplace| Plugin | Description | Version |
|---|---|---|
| cc-suite | CC Suite — one plugin to bridge and delegate across Claude Code, Codex CLI, and Antigravity: single-source AGENTS.md, shared skills, mirrored hooks and MCP servers, full Claude↔Codex bidirectional delegation (including Codex reading Claude's session history), a Claude→Grok Build delegation lane over ACP, a bounded read-only Claude→Qwen Code review lane, project-scoped advisor agents, opt-in bridging to more coding agents (Grok Build, opencode, Qwen Code, Kimi CLI), and a cross-repo sweep that repairs bridge leftovers in every bridged project at once (dead MCP registrations, and skills symlinks left pointing at a pruned plugin version). Supersedes cc-bridge and codex-toolkit. | 2.3.5 |
| tdd-guardian | TDD Guardian — per-lane gates (unit/integration/e2e/contract), spec-strength S1-S6, adversarial spec review, red receipts, per-path critical thresholds, 9-format coverage merging | 0.10.2 |
| echo-sleuth | Echo Sleuth — mine past conversations, manage memory lifecycle, extract knowledge | 0.4.4 |
| loc-guardian | LOC Guardian — enforce per-file pure LOC limits; deterministic counting, per-file ceilings and exemptions, extraction plans measured in pure LOC, and a --check exit-code mode for CI and hooks |
0.4.2 |
| grill | Deep codebase interrogation: a bounded three-finding quick review by default, with explicit full multi-agent reviews, five review styles and eight pressure tests. | 1.4.0 |
| reason-grill | Deep argument interrogation: a bounded quick review by default, with an explicit full adversarial panel and evidence-grounded verdicts. | 0.4.0 |
| docs-guardian | Documentation freshness, accuracy and coverage checks, plus commit and push guards that resolve the target repository and map outgoing source changes to documentation. | 0.3.0 |
| nlpm | NLPM — score, check, fix, and test NL artifacts across Claude Code, Codex CLI, and Antigravity; tier-aware scoring with per-tool overlays (universal nlpm:conventions floor + conventions-claude, conventions-codex, conventions-antigravity overlays); per-tool Hooks tables (the three event vocabularies aren't 1:1 mappable); standalone Python validator (bin/nlpm-check) remains tool-agnostic for pre-commit hooks and CI; manifest-vs-disk consistency check still the differentiator in the Claude ecosystem |
1.4.3 |
| claude-english-buddy | English coaching with configurable on-demand and sampled modes, bounded provider calls, literal-preservation checks and progress reports; the original user prompt remains authoritative. | 0.7.0 |
| mermaid-preview | Mermaid Preview — auto-preview diagrams in the browser on Write/Edit, offline-safe, dark-mode aware; opens the tab once per file | 0.1.6 |
| ui-tokenize | UI Tokenize — block hardcoded UI values; rewrite-first PreToolUse hook corrects literals to design-token references on the way to disk; configurable strict / advisory strictness; per-project surfaces (by file kind) and ignore (by path) exclusion honored by hooks and audit; /ui-tokenize:review dispatches the token-reviewer agent for semantic mis-pick review |
0.4.6 |
| ui-responsive | UI Responsive — advisory responsive-design coach; flags off-catalog breakpoints, bare 100vh, fixed widths without max-width via PostToolUse additionalContext | 0.1.4 |
| north-star | North Star — 260-token system prompt overriding three RLHF-inherited presumptions (independence, calibration, first-principles); ambient + slash command + subagent layered delivery | 0.1.5 |
| mac-it-guy-pro | Personal IT guy for non-technical macOS users — checkups, safe cleanup, undoable file organising, evidence-first diagnosis, backups with a restore drill, home network, private internet access, automation that leaves reusable tools behind, and a tutor grounded in your machine's own numbers; memory that retires stale beliefs by retesting them; 162 test assertions | 1.9.6 |
| eou-foundry | EOU Foundry — recursive governance for Executable Operating Units (EOUs): faceted classification, generating-EOU constraints, ECP-governed change, no-self-approval. 12 skills covering candidate generation, audit, specify, refactor, promote, foundry-wide audit, ECP authoring, init scaffolding, Stage 0 capture, and judgment audit. | 0.8.4 |
| bureau | Bureau — turn AI sessions into a maintained, human-reviewed knowledge base: capture → compile → review → query, gated by trust tiers (proposed → verified → canonical), with a BUREAU.md instruction (imported by CLAUDE.md) that makes every session honor them; renders to an offline gazette (built by the bundled press). A typed review queue (bureau:review) orders the backlog upstream-first and names the action per page; a human applies decisions per-page or as a reviewed, content-bound, commit-gated batch (approve --from / --all). A deterministic recursion engine tracks claim-level dependencies (rests_on + author-anchored spans) so changing an upstream claim flags every downstream page for review; a live Engine view (bureau:serve) shows three signals as you edit — dependency freshness (needs-review/stale), artifact currency (a verified file that drifted out from under a claim), and a convergence trend — and git-backed versioning renders any past board, diffs two versions, and pins named snapshots. Records architecture decisions as MADR ADR pages (bureau:adr) with typed supersedes edges — superseded only once the superseding ADR is approved (ADR-0006). Delegate the review queue to Codex as your representative (bureau:codex-review) — it advises, or, where a workspace opts the codex authority in, commits under --by codex (ADR-0007). Bulk-approve the whole backlog in one confirmation (bureau:approve-all) — the AI prepares and confirms, you fire the single --by human line. Enables a crew of specialized agents. Self-contained. | 1.0.6 |
| xros | XROS — executable research operating system. Four commands: compile a long-horizon investigation into a verifiable methodology spec; sharpen an un-checkable question (orient a newcomer in a field's vocabulary → falsifiable claim → best-available check, with its ceiling); run the Tier-A/B multi-agent engine (diverse routes, adversarial refutation) gated on a real oracle's exit code; or reason through a no-oracle Tier-C claim with premises, pre-mortem, and dated tripwires. Founding rule: no verification, no claim. Ships a bundled dependency-free validator + conformance suite. | 0.3.4 |
| compute-ladder | Compute routing with guarded effort requests, a measured cheap-model worker and reports of observed main-agent and subagent effort; runtime escalation must be verified. | 0.2.0 |
| dev-team | Per-project development roles with separate planner, tester, coder and auditor contexts; validation follows the kind of change, with behavioral changes retaining tests-first gates. | 0.2.0 |
Codex ports are added incrementally as each plugin is converted. Status table — entries are removed from "pending" and added here once their .codex-plugin/ artifacts are committed and smoke-tested.
Most ports keep their skills under codex/skills/. xros instead ships a single root skills/ tree shared by Codex, Google Antigravity, and xAI Grok (all three implement the same skills/<name>/SKILL.md contract), with its .codex-plugin/plugin.json pointing at ./skills/. Antigravity and Grok are installed directly from the plugin repo (agy plugin install <git-url> / grok plugin install xiaolai/xros --trust), not through this marketplace — though Grok also reads Claude Code marketplaces automatically, so grok plugin install xros --trust works once this marketplace is registered.
| Plugin | Description | Version | Status |
|---|---|---|---|
| grill | Deep codebase interrogation: a bounded three-finding quick review by default, with explicit full multi-agent reviews, five review styles and eight pressure tests. | 1.4.0 | Install smoke-tested with Codex CLI 0.144.3 |
| reason-grill | Deep argument interrogation: a bounded quick review by default, with an explicit full adversarial panel and evidence-grounded verdicts. | 0.4.0 | Install smoke-tested with Codex CLI 0.144.3 |
| eou-foundry | EOU Foundry — turn messy workflows into auditable Executable Operating Units with recursive governance (12 skills: candidate generation, audit, specify, refactor, promote, foundry-wide audit, ECP authoring, init scaffolding, Stage 0 capture, judgment audit). v0.7.0 ships Stage 0 (captured_workflow + per-app domain_values constitutional layer + Rule 96 consumption). v0.8.0 ships agentic judgment (judgment_authorized flag, value_invocations trace, F14–F17 taxonomy, judgment_maturity J0–J4 axis, $audit-judgment skill, Rule 97, counterfactual-swap audit). | 0.8.4 | Install smoke-tested with Codex CLI 0.144.3 |
| nlpm | NLPM — reference-knowledge skills (17): the 50 Rules, the 100-point scoring rubric, per-tool conventions (Claude Code / Codex CLI / Antigravity), anti-patterns, vocabulary discipline, and authoring guides. Knowledge skills, not commands — $nlpm-rules + $nlpm-scoring to score an artifact, $nlpm-conventions-codex for the Codex layout. The interactive /nlpm:* linting commands stay a Claude Code plugin (they orchestrate sub-agents); cross-tool deterministic checks use the standalone bin/nlpm-check. |
1.4.3 | Install smoke-tested with Codex CLI 0.159.2 (GitHub source, temporary CODEX_HOME): 17 skills, no source-command-* duplicates, no hooks; command/agent orchestration stays Claude-only |
| xros | XROS — an eXecutable Research Operating System. Four skills ($xros-compile, $xros-sharpen, $xros-run, $xros-reason): turn an investigation into a verifiable methodology spec, then gate the verdict on a real oracle's exit code. Founding rule: no verification, no claim. Ships a dependency-free validator + conformance suite. |
0.3.4 | Install smoke-tested with Codex CLI 0.144.3 |
| mac-it-guy-pro | Personal IT guy for non-technical macOS users — checkups, safe cleanup, undoable file organising, evidence-first diagnosis, backups with a restore drill, home network, automation that leaves reusable tools behind, and a tutor grounded in your machine's own numbers. PreToolUse guard and SessionStart digest both run under Codex. macOS only. | 1.9.6 | Install smoke-tested with Codex CLI 0.159.2 (GitHub source, temporary CODEX_HOME; 8 skills and both hooks listed); hook firing inside a live session not yet exercised |
| tdd-guardian | TDD Guardian — per-lane gates, spec-strength S1-S6, adversarial spec review, red receipts, critical-path thresholds, coverage merging and mutation testing as 11 workflow skills ($tdd-guardian-workflow, $tdd-guardian-gate, ...). PreToolUse commit/push freshness hook runs under Codex; Claude Code's TaskCompleted auto-gate has no Codex event, so $tdd-guardian-gate runs those lanes. |
0.10.2 | Install smoke-tested with Codex CLI 0.159.2 (GitHub source, temporary CODEX_HOME; skill listing checked); hook firing inside a live session not yet exercised |
| loc-guardian | LOC Guardian — $loc-guardian-scan counts pure LOC through a tested reducer (no model arithmetic), flags files over the limit, and proposes extractions measured with tokei; $loc-guardian-init sets the limit and rules. |
0.4.2 | Install smoke-tested with Codex CLI 0.159.2 (GitHub source, temporary CODEX_HOME; skill listing checked) |
| docs-guardian | Documentation freshness, accuracy and coverage checks, plus commit and push guards that resolve the target repository and map outgoing source changes to documentation. | 0.3.0 | Install smoke-tested with Codex CLI 0.159.2 (GitHub source, temporary CODEX_HOME; skill listing checked); hook firing inside a live session not yet exercised |
| claude-english-buddy | English coaching with configurable on-demand and sampled modes, bounded provider calls, literal-preservation checks and progress reports; the original user prompt remains authoritative. | 0.7.0 | Smoke-tested with Codex CLI 0.159.2 (GitHub source, temporary CODEX_HOME): 17 skills listed, the hook corrected a prompt and logged it, and the report skills read that history; trust was bypassed for the run, and systemMessage display in the TUI is unverified |
Pending ports: none. echo-sleuth is not ported: it mines and manages Claude Code's own transcripts and memory files, so under Codex it would be blind to Codex's sessions, and cc-suite already gives Codex read access to Claude history. The Claude↔Codex delegation lane is now handled by cc-suite (single plugin, bidirectional). codex-guardian (Codex-artifact auditor) remains a planned future addition for the Codex-only audit niche.
claude plugin install cc-suite@xiaolai --scope user
claude plugin install tdd-guardian@xiaolai --scope user
claude plugin install echo-sleuth@xiaolai --scope user
claude plugin install loc-guardian@xiaolai --scope user
claude plugin install grill@xiaolai --scope user
claude plugin install docs-guardian@xiaolai --scope userOne-time setup (shell):
codex plugin marketplace add xiaolai/claude-plugin-marketplaceCodex registers this as marketplace name xiaolai.
Per-plugin install (shell):
codex plugin list --marketplace xiaolai --available
codex plugin add grill@xiaolaiYou can also install from inside a Codex session:
- Start a session:
codex(in your project directory) - Type
/plugins— opens the plugin TUI - Find grill in the xiaolai marketplace, install/enable it
Invoking a plugin (inside a Codex session):
Codex uses the skill prefix $, not slash commands. Type:
$grill-roast
Or just describe the task in natural language ("do a multi-angle audit of this codebase") and Codex's auto-match will load the skill from its description.
(More entries to come as Codex ports land.)
claude plugin install cc-suite@xiaolai --scope project
claude plugin install tdd-guardian@xiaolai --scope project
claude plugin install echo-sleuth@xiaolai --scope project
claude plugin install loc-guardian@xiaolai --scope project
claude plugin install grill@xiaolai --scope project
claude plugin install docs-guardian@xiaolai --scope project| Scope | Flag | Effect |
|---|---|---|
| User (global) | --scope user |
Available in all projects (default) |
| Project | --scope project |
Shared with team via .claude/plugins.json |
| Local | --scope local |
Local only, not committed to git |
claude plugin list # List installed plugins
claude plugin update grill@xiaolai # Update to latest version
claude plugin disable grill@xiaolai # Temporarily disable
claude plugin enable grill@xiaolai # Re-enable
claude plugin uninstall grill@xiaolai # RemoveThe marketplace is a local git clone, and claude plugin install does not auto-refresh it before resolving the plugin name. If a plugin was added to the marketplace after your local clone was last updated, install will fail with a misleading "not found" error.
Refresh the marketplace, then retry:
claude plugin marketplace update xiaolai
claude plugin install <plugin-name>@xiaolai --scope userThis is a Claude Code CLI limitation, not a marketplace configuration issue. The plugin is genuinely listed in marketplace.json; your local copy is just stale.
python3 scripts/plugin-health.py /path/to/claude-plugins inventories source versions, cached installations, scope and initialization evidence. It is read-only. Registration does not prove a plugin loaded or that its gate was exercised; those require a fresh runtime load and behavior checks.