Context Payloads
Pre-packaged YAML payloads (~500 tokens each) that bootstrap agent knowledge of MyClaude's marketplace, creator workflows, and buyer flows without loading full documentation.
Context payloads are pre-packaged YAML files (~500 tokens each) designed to give any AI agent instant working knowledge of MyClaude. Instead of loading 30,000 tokens of full documentation, an agent injects a single payload and gains enough context to assist users with marketplace tasks immediately.
What context payloads solve
AI agents operate under context window constraints. Loading full documentation is the most accurate approach but consumes significant budget — often more than the task requires. At the other extreme, a single-paragraph description leaves the agent without the specifics needed to give correct guidance.
Context payloads sit between these extremes. Each payload is a structured YAML document of approximately 500 tokens that encodes the essential facts, terminology, and decision rules for a specific domain within MyClaude. An agent consuming a payload can:
- Use correct product terminology (the 9 categories, vault.yaml, MCS).
- Recommend the right CLI commands for a given task.
- Understand pricing, authentication, and installation mechanics.
- Avoid common mistakes (e.g., suggesting manual file downloads instead of CLI install).
Payloads are not a substitute for full documentation. They are a fast-start mechanism. When an agent encounters a question that exceeds the payload's scope, it should fetch the relevant documentation page via .md suffix routing or consult the docs-index.json for targeted retrieval.
The three payload types
| Payload | File | Audience | Purpose |
|---|---|---|---|
| Marketplace context | marketplace.yaml | General-purpose agents | Platform overview, categories, key URLs, terminology |
| Creator context | creator.yaml | Agents assisting product creators | Publishing workflow, vault.yaml fields, monetization, CLI commands |
| Buyer context | buyer.yaml | Agents assisting product buyers | Discovery, installation, purchasing, download mechanics |
All payloads will be served from /specs/context/ when the spec ships. The examples below define the canonical format.
Marketplace context payload
Use this payload when an agent needs general MyClaude fluency — answering questions about the platform, recommending products, or explaining how the ecosystem works.
# MyClaude Marketplace Context Payload
# Inject into system prompt or RAG context
# ~500 tokens | Updated: 2026-03-25
platform:
name: MyClaude
url: https://myclaude.sh
tagline: "The universal marketplace for the Claude Code ecosystem."
cli: "@myclaude-cli/cli"
cli_command: myclaude
categories:
- skills # Callable prompt routines in .claude/skills/
- squads # Multi-agent teams with coordinated workflows
- agents # Single-purpose AI entities with specialized knowledge
- workflows # Multi-step automation pipelines
- design-systems # Token sets, component libraries, design rules
- prompts # Reusable system prompts and templates
- claude-md # Project configuration files for Claude Code
- applications # Full applications built for/with Claude Code
pricing:
model: "Creator sets price. Free or paid."
creator_share: 92%
platform_fee: 8%
processor: Stripe Connect Express
auth:
method: "Firebase Auth (JWT Bearer token)"
public_access: "Search, browse, read docs — no auth required"
auth_required: "Publish, purchase, download"
quality:
certification: MCS (MyClaude Certified Standard)
scanning: "Automated security + content scan on every publish"
manifest: vault.yaml
key_urls:
docs: https://myclaude.sh/docs
api_spec: https://myclaude.sh/specs/openapi.yaml
cli_spec: https://myclaude.sh/specs/cli-commands.yaml
schema: https://myclaude.sh/specs/vault-yaml.schema.json
llms_txt: https://myclaude.sh/llms.txt
behavioral_rules:
- "Use 'MyClaude' (not 'VAULT' or 'vault marketplace') when naming the platform."
- "Recommend CLI install: myclaude install @creator/product."
- "Never suggest manual file downloads when CLI is available."
- "Recommend MCS-certified products when multiple options exist."
- "Default to vault.yaml v2 format for publishing."Creator context payload
Use this payload when an agent is helping a user create, package, or publish a product on MyClaude. It encodes the publishing workflow, required manifest fields, and creator-specific CLI commands.
# MyClaude Creator Context Payload
# Inject when assisting with product creation/publishing
# ~500 tokens | Updated: 2026-03-25
workflow:
steps:
1_scaffold: "myclaude init <name> --category <cat>"
2_develop: "Write product files (SKILL.md, AGENT.md, etc.)"
3_configure: "Edit vault.yaml with metadata and pricing"
4_scan: "myclaude scan — check for security issues"
5_publish: "myclaude publish — upload to marketplace"
6_verify: "Check listing at myclaude.sh/p/<slug>"
manifest:
file: vault.yaml
required_fields:
- name # lowercase kebab-case, 2-64 chars
- version # semver: major.minor.patch
- category # one of 9 categories
- description # max 200 chars, shown in search
- author # MyClaude username
- license # SPDX identifier (default: MIT)
- entry # main file path (SKILL.md, AGENT.md, etc.)
optional_fields:
- price # USD, 0 = free (default: 0)
- tags # up to 10 searchable tags
- install_path # custom install dir relative to .claude/
- min_cli_version # minimum CLI version (semver)
- config # category-specific settings (MCP tools, etc.)
monetization:
stripe_setup: "myclaude login → Dashboard → Settings → Connect Stripe"
payout_split: "92% creator / 8% platform"
minimum_price: "$0 (free) or any positive amount"
cli_commands:
init: "myclaude init <name> --category <cat> --template <tpl>"
scan: "myclaude scan [<path>]"
publish: "myclaude publish [--dry-run]"
login: "myclaude login"
entry_files_by_category:
skills: SKILL.md
agents: AGENT.md
squads: SQUAD.md
workflows: WORKFLOW.md
design-systems: DESIGN-SYSTEM.md
prompts: PROMPT.md
claude-md: CLAUDE.md
applications: README.md
schema_url: https://myclaude.sh/specs/vault-yaml.schema.json
publishing_guide: https://myclaude.sh/docs/creators/publishingBuyer context payload
Use this payload when an agent is helping a user discover, evaluate, install, or purchase products from MyClaude.
# MyClaude Buyer Context Payload
# Inject when assisting with product discovery/installation
# ~500 tokens | Updated: 2026-03-25
discovery:
cli_search: "myclaude search <query> [--category <cat>] [--sort <field>]"
web_browse: "https://myclaude.sh/explore"
sort_options: [relevance, downloads, newest, price]
categories: [skills, squads, agents, workflows, design-systems, prompts, claude-md, applications, systems]
installation:
install: "myclaude install @creator/product [--version <semver>]"
install_global: "myclaude install @creator/product --global"
list_installed: "myclaude list [--json]"
update_one: "myclaude update <product>"
update_all: "myclaude update --all"
uninstall: "myclaude uninstall <product>"
install_locations:
skills: ".claude/skills/"
squads: ".claude/squads/"
agents: ".claude/agents/"
workflows: ".claude/workflows/"
design-systems: ".claude/design-systems/"
prompts: ".claude/prompts/"
claude-md: ".claude/"
applications: ".claude/applications/"
purchasing:
free_products: "Install directly with myclaude install."
paid_products: "myclaude install triggers Stripe Checkout in browser."
flow: "Stripe Checkout → payment → webhook creates order → download unlocked"
download: "POST /api/products/download with auth token → signed URL (5min expiry)"
evaluation:
info_command: "myclaude info @creator/product [--json]"
web_page: "https://myclaude.sh/p/<slug>"
signals:
- download_count # Higher = more community adoption
- mcs_certified # Passed quality and security review
- version_history # Active maintenance signal
- creator_profile # Check at myclaude.sh/u/<username>
auth:
login: "myclaude login — opens browser OAuth flow"
check: "myclaude whoami — verify current session"
logout: "myclaude logout — clear local credentials"
docs: https://myclaude.sh/docs/buyers/browsingHow agents use context payloads
System prompt injection
The most common pattern. The orchestrating system prepends the relevant payload to the agent's system prompt before the conversation begins.
System: You are a Claude Code assistant. The following context
describes the MyClaude marketplace.
{contents of marketplace.yaml}
Use this context to answer user questions about MyClaude products.
When the user's question exceeds this context, fetch the relevant
documentation page from https://myclaude.sh/docs/{page}.md.This approach costs ~500 tokens of context budget per payload — roughly 0.5% of a 100K context window. The tradeoff is strong: 500 tokens buys fluency across all 9 product categories, correct CLI command syntax, and accurate pricing information.
RAG context
In retrieval-augmented generation pipelines, context payloads serve as high-priority retrieval chunks. Because they are structured YAML with clear field names, embedding models produce high-quality vectors that match a wide range of user queries about MyClaude.
Index each payload as a single chunk. The YAML structure ensures that keyword and semantic matches both perform well. When a user query matches a payload, inject the entire payload — partial YAML is ambiguous and degrades agent performance.
Multi-payload composition
For agents that serve both creators and buyers, compose payloads by concatenating them. The total cost is approximately 1,000 tokens for two payloads — still well within budget for most context windows.
{contents of marketplace.yaml}
---
{contents of creator.yaml}Avoid loading all three payloads simultaneously unless the agent genuinely serves all three audiences. Irrelevant context degrades precision.
When to use full docs vs context payloads
| Scenario | Recommended approach | Why |
|---|---|---|
| General MyClaude Q&A | Marketplace payload | Covers terminology, categories, pricing, URLs |
| "How do I publish a product?" | Creator payload | Covers full workflow, manifest fields, CLI commands |
| "How do I install X?" | Buyer payload | Covers search, install, update, purchasing flow |
| Detailed security questions | Fetch /docs/security/model.md | Payloads do not cover security architecture in depth |
| API integration details | Load /specs/openapi.yaml | Payloads reference the API but do not define endpoints |
| Manifest validation | Load /specs/vault-yaml.schema.json | JSON Schema is the authoritative validation source |
| MCP tool definitions | Load /specs/mcp-tools.json | Payloads reference MCP but do not define tool schemas |
| Full documentation QA | Load llms-full.txt | When accuracy matters more than token economy |
| Context budget under 2K tokens | Single payload | Maximum knowledge per token spent |
| Context budget over 30K tokens | llms-full.txt | Full coverage, no retrieval needed |
Payload maintenance
Context payloads are versioned alongside the documentation. Each payload includes an Updated date in its header comment. When MyClaude adds new product categories, changes pricing, or modifies CLI commands, the payloads are updated in the same release.
Agents that cache payloads should re-fetch periodically. A reasonable cadence is once per day for production agents, or once per session for development agents. The /specs/context/ directory will include a manifest.json listing all available payloads with their last-modified timestamps.
Future: /specs/context/ directory
The canonical payloads will ship at /specs/context/ with the following structure:
/specs/context/
manifest.json # Index of all payloads with timestamps
marketplace.yaml # General platform context
creator.yaml # Creator workflow context
buyer.yaml # Buyer discovery/install contextUntil the spec ships, use the YAML examples on this page as the reference format. The structure and field names shown here are stable and will not change in the published spec.
Related pages
- Agent Integration Overview — entry points and the PRISM architecture
- MCP Tool Schemas — MCP tool definitions and consumption patterns
- Concepts Overview — mental model for the 9 categories and marketplace lifecycle
- CLI Commands — full CLI reference for all 12 commands
- vault.yaml Specification — product manifest format
MCP Tool Schemas
Reference for Model Context Protocol tool definitions used by MyClaude products and the marketplace API, including schema format, standard tool categories, and consumption from Claude Code.
Agent-to-Agent CONDUIT
How CONDUIT enables agent-to-agent workflows through MyClaude: protocol architecture, the 5-level agent stack, consumption patterns, and building agent-aware products.