CCAM Skills and Plugin Marketplace

August 10, 2026 · View on GitHub

CCAM ships one source tree that supports three distribution paths:

Claude Code plugins through .claude-plugin/marketplace.json
Codex plugins through .agents/plugins/marketplace.json and each plugin's .codex-plugin/plugin.json
Open Agent Skills / skills.sh through standards-compliant SKILL.md files with required name and description metadata

The verified bundle contains 14 plugins, 66 bundled plugin skills, 18 Claude subagents, 34 Claude commands, 3 CLI helpers, 3 hook configurations, and 2 MCP-enabled plugins. The skills.sh CLI discovers 75 total repository skills because it also includes the repository-maintenance skills under .agents/skills/ and .claude/skills/.

Choose an Installation Path

Need	Recommended path	Why
A curated capability pack for Claude Code	Claude Code plugin marketplace	Installs the plugin's skills, commands, agents, and metadata together.
The same pack for Codex	Codex plugin marketplace	Uses the Codex manifest while sharing the canonical plugin source tree.
One reusable workflow without the rest of a plugin	`npx skills add`	Keeps the install focused on the selected skill and supports project or global scope.
Modify or contribute an extension	Clone this repository	Lets you run the sync and validation commands before using local manifests.

Do not install the same capability through multiple paths unless you intentionally want duplicates. Start with one path, verify it with the listed CLI command, and switch only after removing the previous installation.

Install for Claude Code

claude plugin marketplace add hoangsonww/Claude-Code-Agent-Monitor
claude plugin install ccam-platform@claude-code-agent-monitor-plugins

List the marketplace and install any plugin shown in the catalog:

claude plugin marketplace list
claude plugin validate . --strict

Install for Codex

codex plugin marketplace add hoangsonww/Claude-Code-Agent-Monitor
codex plugin list --marketplace claude-code-agent-monitor-plugins --available --json
codex plugin add ccam-platform@claude-code-agent-monitor-plugins

Codex reads .agents/plugins/marketplace.json. Every entry points to the same plugins/<name> directory used by Claude Code. Each plugin has a separate .codex-plugin/plugin.json with UI metadata, capabilities, skills, and optional MCP wiring.

Install skills with skills.sh

The repository does not need a PR to the vercel-labs/skills source repository. The skills CLI installs from public GitHub repositories directly, and skills.sh discovery/ranking is driven by install telemetry.

# List all 75 repository skills without installing
npx skills add hoangsonww/Claude-Code-Agent-Monitor --list

# Install one skill into the current project for Claude Code and Codex
npx skills add hoangsonww/Claude-Code-Agent-Monitor \
  --skill mcp-server \
  --agent claude-code \
  --agent codex \
  --yes

# Verify project-scoped installation
npx skills list --json

# Install the same skill globally for both agents
npx skills add hoangsonww/Claude-Code-Agent-Monitor \
  --skill mcp-server \
  --agent claude-code \
  --agent codex \
  --global \
  --yes

# Verify global installation
npx skills list --global --json

# Install all discovered skills for every supported agent in the current project
npx skills add hoangsonww/Claude-Code-Agent-Monitor --all

# Update project-scoped or global skills
npx skills update --project --yes
npx skills update --global --yes

# Remove the selected skill from the current project or global scope
npx skills remove mcp-server --yes
npx skills remove --global mcp-server --yes

Project installs use the current repository's .agents/skills/ directory and create agent-specific links such as .claude/skills/mcp-server. By default, global Claude Code skills resolve under ~/.claude/skills/ and global Codex skills under ~/.codex/skills/; CLAUDE_CONFIG_DIR and CODEX_HOME replace those respective base directories. Multi-agent installs may deduplicate files through a shared store and link the agent-specific destinations. The CLI records source and hash metadata in skills-lock.json for project installs and a global skill lockfile so later skills update runs can check the original GitHub source.

Every bundled skill carries:

SKILL.md frontmatter with canonical name and description
procedural instructions and safety boundaries
agents/openai.yaml with Codex/ChatGPT UI metadata and a $skill-name default prompt

The public repository becomes installable as soon as these files are on the default branch. Site search or leaderboard visibility can lag and depends on real installs. This change does not publish or submit anything externally.

Plugin Catalog

Plugin	Focus	Bundled skills
`ccam-analytics`	tokens, costs, cache, model mix, trends	`cache-efficiency`, `cost-breakdown`, `model-mix`, `productivity-score`, `session-report`, `usage-trends`
`ccam-config`	Claude config and memory governance	`config-audit`, `hook-inventory`, `mcp-audit`, `memory-review`, `skill-inventory`
`ccam-cost-guard`	budgets, forecasts, cost alerts	`budget-set`, `cost-alert`, `daily-budget-check`, `model-savings`, `spend-forecast`
`ccam-dashboard`	dashboard status and MCP connector	`dashboard-status`, `endpoint-probe`, `live-watch`, `quick-stats`
`ccam-devtools`	diagnostics, export, transcripts	`data-export`, `event-trace`, `health-check`, `hook-diagnostics`, `session-debug`, `transcript-grep`
`ccam-insights`	anomalies, patterns, regressions	`anomaly-alert`, `benchmark`, `optimization-suggest`, `pattern-detect`, `regression-watch`, `session-compare`
`ccam-integrations`	alerts, webhooks, remote collection	`alert-management`, `remote-collection`, `webhook-management`
`ccam-platform`	Claude/Codex config, hooks, import, backup, MCP	`config-explorer`, `history-portability`, `hook-setup`, `mcp-server`
`ccam-productivity`	standups and reviews	`daily-standup`, `monthly-review`, `sprint-summary`, `time-of-day`, `weekly-report`, `workflow-optimizer`
`ccam-quality`	errors, SLOs, hook reliability	`api-error-report`, `error-scan`, `hook-failure-audit`, `regression-alert`, `slo-check`
`ccam-reports`	executive, cost, reliability, workflow reports	`cost-report`, `executive-report`, `reliability-report`, `workflow-report`
`ccam-runner`	monitored Claude Code/Codex launch and history	`run-agent`, `run-history`
`ccam-sessions`	session search, timeline, replay, cleanup	`cwd-rollup`, `session-cleanup`, `session-search`, `session-timeline`, `transcript-replay`
`ccam-workflows`	orchestration and fleet intelligence	`concurrency-report`, `dag-map`, `delegation-audit`, `error-propagation`, `fleet-runs`

MCP-enabled plugins

ccam-dashboard and ccam-platform include:

{
  "mcpServers": {
    "ccam-dashboard": {
      "command": "ccam",
      "args": ["mcp", "stdio"]
    }
  }
}

The stable ccam mcp stdio launcher resolves the MCP build from the linked CCAM checkout. This avoids plugin-cache-relative paths, which break after installation. Run npm run setup first. It now installs and builds MCP before linking ccam.

Source layout

.claude-plugin/
└── marketplace.json                 # Claude Code catalog
.agents/plugins/
└── marketplace.json                 # Codex catalog
plugins/<name>/
├── .claude-plugin/plugin.json       # Claude plugin manifest
├── .codex-plugin/plugin.json        # Codex plugin manifest
├── skills/<skill>/
│   ├── SKILL.md
│   └── agents/openai.yaml
├── agents/*.md                      # Claude subagents where applicable
├── commands/*.md                    # Claude commands where applicable
├── hooks/hooks.json                 # optional
├── bin/*                            # optional CLI helpers
└── .mcp.json                        # optional MCP server wiring

node scripts/sync-agent-extensions.js is the deterministic source synchronizer. It:

adds missing canonical skill names
writes agents/openai.yaml
generates .codex-plugin/plugin.json
rebuilds both marketplace catalogs
prints verified component totals

Do not hand-edit generated Codex manifests or skill agents/openai.yaml. Edit the Claude manifest or SKILL.md, then run:

npm run extensions:sync
npm run extensions:validate

Validation

npm run extensions:sync
npm run extensions:validate
node --test server/__tests__/plugins-marketplace.test.js
claude plugin validate . --strict
npx skills add . --list

For a clean Codex installation test:

tmp="$(mktemp -d)"
CODEX_HOME="$tmp" codex plugin marketplace add "$PWD" --json
CODEX_HOME="$tmp" codex plugin add ccam-platform@claude-code-agent-monitor-plugins --json
CODEX_HOME="$tmp" codex plugin list --json
rm -rf "$tmp"

The repository test enforces:

marketplace-to-directory bijection for Claude and Codex
valid dual manifests whose names match their folders
required policy/category fields in the Codex catalog
skill name-to-directory agreement
description in every skill
agents/openai.yaml in every skill with a $skill-name prompt
required frontmatter for Claude agents and commands
valid hook JSON

Safety

Most analytical plugins are read-only.
Skills that can mutate show the intended change and require confirmation.
Webhook tests and push notifications are external sends.
Run tools launch or control real local processes.
Config edits remain inside server allowlists and create backups.
Remote-source deletion retains data unless purge is explicitly confirmed.
Full dashboard clearing stays behind MCP mutation/destructive flags and the exact CLEAR_ALL_DATA token.

Public publication

A repository marketplace is immediately usable by Claude Code and Codex from Git. Public inclusion in OpenAI's universal plugin directory is a separate reviewed submission through OpenAI's plugin submission portal. This repository adds publication-ready manifests but does not submit them or perform any external publication action.