The rules the reference agents enforce in code for third-party content, writes, grounding, identity, and memory, each with its module, plus two adversarial-eval rules. Load when handling untrusted tool results, guarding a write tool, or scoping what an agent remembers.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
commerce_common/ is commerce-common/commerce_common/, shopping_agent/ is shopping-agent/core/shopping_agent/, and
merchant_agent/ is merchant-agent/core/merchant_agent/; docs/safety.md lists the same rules. A rule inside a tool
call holds on all three paths, which share one executor (commerce-architecture); rules 10 and 15 name the paths they run
on. Merchant staging, guardrails, approval, and marketplace posture are in commerce-merchant-operations.
Fence.fence_payload (commerce_common/fencing.py): NFKC-normalized, invisible and control characters removed, fence
markers, forged turn markers, and special tokens replaced, wrapped in the role's label, and cut at max_fenced_chars.
The executor's _fenced applies it to every handler.STOREFRONT_FENCE in shopping_agent/fencing.py, MERCHANT_FENCE in
merchant_agent/fencing.py); the notice appears once, in the static prompt, with nothing untrusted in it.build_dynamic_context in each role's prompt.py); backend context blocks have their own cap.max_search_results (clamp_limit in commerce_common/execution.py).ShoppingSessionState.seen_products, filled by the session's catalog and order reads
(check_provenance and remember_order_items in shopping_agent/gates.py); update and remove also accept a line
already in the cart. Merchant staging accepts only ids in seen_listings and read_listings, apply and discard only
ids in seen_changes (merchant_agent/gates.py; commerce-merchant-operations).max_quantity_per_item caps the line after the write and max_cart_lines the cart, under a session lock
(gated_add_to_cart in shopping_agent/gates.py; caps in shopping_agent/config.py). The lock is per process, so
StorefrontBackend cart methods enforce them too; eligibility, pricing, and inventory are the backend's in both roles.checkout renders the cart (enrich_checkout in
shopping_agent/enrichment.py) and the host completes it; post-purchase care is reads, with no cancel or refund tool.ShoppingSessionState, MerchantSessionState), saved with the session when
the request or turn ends (SessionStore in examples/demo_common/sessions.py, a versioned state document beside the
transcript), so a request on another process loads it. Each map keeps its newest PROVENANCE_CAP records (remember
in commerce_common/types.py); a dropped id needs a fresh read.run_presentation in commerce_common/presentation.py; each role's
enrichment.py; commerce-ui-tools).GROUNDING_RULES in shopping_agent/grounding.py), or
a performance question or an apply request with nothing staged (merchant_agent/grounding.py), starts from the
matching read: first_forced_tool in commerce_common/grounding.py picks it, the Messages API runtimes force it with
tool_choice, and the SDK runtimes prefetch it when the rule has a prefetch form (ground in
commerce_common/agent_sdk.py; the terms rule has none); the hosted path has the prompt only.policy_intent_terms, order_intent_terms, product_id_patterns, metrics_intent_terms,
and their cues); a deployment appends its own words, a gate flag turns a rule off, and neither changes prompt bytes.SessionStore.start and session_dependency in
examples/demo_common/sessions.py). No request field or tool argument names a user, merchant, or operator; the MCP
servers take the principal from their environment, a production server from its request.save_memory and post-turn extraction) passes validate_fact in
commerce_common/memory.py: key of at most 64 characters, value of at most 200, one of the three MemoryCategory
values, and the MemoryWriteFilter, which refuses identifier-shaped values by default; memory_blocked_patterns
adds patterns, and a filter with checks replaces it (MemoryRuntime.build).transcript_text in commerce_common/turn.py), and
extract_and_store drops its batch when the subject was purged meanwhile; the Messages API runtimes run it
(update_memory), the SDK host calls the runtime, and the hosted path writes through save_memory only.memory_retention_days (with_retention), MemoryStore.delete_fact, MemoryStore.clear, and enable_memory hold
on every path without changing prompt or tool bytes; the examples expose read and delete routes
(install_memory_routes in examples/demo_common/memory.py), and clear belongs in account deletion.user_id or the operation's merchant_id (memory_subject in each role's executor.py).ToolOutcome.held in commerce_common/streaming.py; the host's
tool_result event carries status: blocked), a failure returns an error result, and execute never raises.enable_web_search (default off) adds the tool, the SDK runtimes allow-list
the registered names, and the manifests enable tools one by one.enforce_local_only_bind in commerce_common/mcp_server.py).まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Creating and improving listing content, covering titles, descriptions, attribute completeness, categorization fixes, image callouts written as text, edits written from material the operator supplies, and content-quality audits and bulk fixes across many listings. Not needed for questions about how a listing is performing.
日本語の概要は準備中です。原文の説明を表示しています。
How the reference commerce agents of either role are structured, covering the loop, where each rule lives, skills, the backend interface, delegates, and model fields. Load when designing or reviewing the structure of a shopping or merchant agent.
日本語の概要は準備中です。原文の説明を表示しています。
Authoring and running behavioral evals for a shopping or merchant agent, covering the case shape, authoring rules, code graders and judges, the run pattern, and poisoned fixtures. Load when writing eval cases or rubrics or deciding how a suite runs.
日本語の概要は準備中です。原文の説明を表示しています。
The reference merchant agent, covering its flows, staged changes and host approval, metrics grounding, the analysis delegate, store memory, components, and marketplace rules. Load when building or reviewing an agent for the operators of a store, property, subscriber base, or venue.
日本語の概要は準備中です。原文の説明を表示しています。
The reference agents' cache-stable request assembly, covering the static system and per-request context split, the fixed tool list, the rolling conversation breakpoint, which config fields are prompt bytes, and verification. Load when writing or reviewing a commerce agent's prompt assembly, when cache reads are zero, or when a turn's latency or cost is the question.
日本語の概要は準備中です。原文の説明を表示しています。
The reference presentation-tool contract, covering server-side enrichment, suggestion chips, the event stream, progressive rendering, both roles' built-in components, and adding a vertical component. Load when building or reviewing how a commerce agent's output reaches a frontend.
日本語の概要は準備中です。原文の説明を表示しています。