asm-protocol 0.5.2__tar.gz → 0.6.0__tar.gz

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: asm-protocol
-Version: 0.5.2
+Version: 0.6.0
 Summary: Agent Service Manifest: value metadata and scoring for autonomous service selection
 Author: Yi Guo
 License-Expression: MIT
@@ -41,7 +41,7 @@ It accepts either a standalone ASM manifest or an MCP Registry `server.json`
 with publisher-provided ASM metadata:
 
 ```bash
-python -m pip install "asm-protocol==0.5.2"
+python -m pip install "asm-protocol==0.6.0"
 asm-lint server.json --format markdown --output asm-lint-report.md
 ```
 
@@ -58,7 +58,7 @@ steps:
   - uses: actions/setup-python@v5
     with:
       python-version: "3.12"
-  - uses: YE-YI7/asm-spec/.github/actions/asm-lint@v0.5.2
+  - uses: YE-YI7/asm-spec/.github/actions/asm-lint@v0.6.0
     with:
       path: server.json
       fail-on: invalid
@@ -67,6 +67,8 @@ steps:
 The Action adds the full Markdown report to the job summary. It does not call
 an ASM-hosted API or upload the inspected manifest. See the
 [lint and CI guide](docs/adoption/asm-lint.md) for status semantics.
+For a bounded producer-side change, use the
+[10-minute adoption package](docs/adoption/ten-minute.md).
 
 ## Try it: pick a tool for a task
 
@@ -75,7 +77,13 @@ git clone https://github.com/YE-YI7/asm-spec.git && cd asm-spec
 python library/select_demo.py
 ```
 
-For *"make a study plan and remind me daily"* with a cloud agent on Windows, the selector drops the tools it can't drive (Apple Reminders, Things 3 — local-device only) and the ones it can't call directly (Any.do — Zapier only), then ranks the rest. Ask for a built-in pomodoro and the pick changes to TickTick. Ask to *"edit an image and lay out a poster"* and it picks free, scriptable **Photopea** over paid Photoshop — and filters **Affinity Designer**, which exposes no automation API at all.
+The deterministic core does **not** pretend to understand the task sentence. The
+caller supplies structured facts such as `taxonomy`, `required_functions`,
+platform, reach, and expected workload; `task` remains audit/display text. If
+taxonomy and required functions are both absent, selection returns
+`under_specified` instead of choosing an unrelated globally cheap tool.
+
+For *"make a study plan and remind me daily"* with a cloud agent on Windows, the selector drops the tools it can't drive (Apple Reminders, Things 3 — local-device only) and the ones it can't call directly (Any.do — Zapier only), then ranks the rest from those explicit constraints. Ask for a built-in pomodoro and the pick changes to TickTick. Ask to *"edit an image and lay out a poster"* and it filters **Affinity Designer**, which exposes no automation API at all.
 
 The library it selects over is in [`library/`](library/) — 30 real tools across task management, creative design, research, communication, developer tools, booking, and real-estate data today, each carrying:
 
@@ -105,31 +113,50 @@ ASM is MCP-compatible: publish a standalone `.well-known/asm`, or embed ASM in M
 ASM ships an MCP server so any MCP client (Claude Desktop, Cursor, an agent host) can call the selector as a tool — no schema adoption required:
 
 ```bash
-python3 -m pip install "asm-protocol[mcp]==0.5.2"
+python3 -m pip install "asm-protocol[mcp]==0.6.0"
 asm-selector                     # stdio MCP server (MCP SDK 2.x)
 ```
 
 It exposes three tools: **`select_tool`** (pick a tool for a task and return its risk/approval policy), **`list_library_tools`**, and **`get_tool_manifest`**. Point your client's MCP config at `asm-selector`, or at `python3 /path/to/asm-spec/asm_selector_mcp.py`; the selector reads `library/` (override with `ASM_LIBRARY_DIR`). The Python server uses the stable MCP SDK 2.x line and supports the modern `2026-07-28` protocol era. The same selector is importable directly: `from library_select import select`.
 
+Cost output has an explicit `known`, `partial`, or `unknown` status. Metered
+prices need expected monthly usage; one-time licenses need an amortization
+period; prose-only free tiers remain unknown because their allowance and reset
+rules are not machine-readable. If every eligible candidate does not have a
+known cost in the same currency, the selector returns `needs_cost_facts` rather
+than guessing. A caller may explicitly request the `capability_breadth` fallback;
+it is never applied implicitly.
+
 The same engine is also available as:
 
 ```bash
 # CLI (human or scripted)
 asm select "find and book a refundable flight" --taxonomy tool.booking.travel \
-  --requires flight_search,flight_order_create --json
+  --requires flight_search,flight_order_create --fallback-policy capability_breadth --json
 
 # Hosted HTTP API (stdlib-only; deploy anywhere that runs Python)
 python asm_select_api.py     # POST /select, GET /tools, GET /healthz on :8787
 ```
 
-LangChain / LangGraph builders get the same selector as a drop-in tool (`pip install langchain-core`):
+DeepSeek Harness developer-preview users can install the native
+[`asm_select` tool adapter](integrations/deepseek-harness/README.md). It uses
+the same HTTP contract and returns the current structured decision, but never
+invokes or authorizes the selected service. The frozen v0.1 receipt is available
+only through an explicit legacy compatibility profile. The adapter defaults to
+a local selector so task text is not sent to a hosted endpoint implicitly.
+
+LangChain / LangGraph builders get the same selector as a packaged drop-in tool:
 
 ```python
-import sys; sys.path += ["asm-spec", "asm-spec/integrations/langchain"]
-from asm_tools import ASMToolSelectorTool
+from asm_protocol.integrations.langchain import ASMToolSelectorTool
 agent_tools = [ASMToolSelectorTool()]   # name: asm_tool_selector
 ```
 
+Install it with `python -m pip install "asm-protocol[langchain]==0.6.0"`. LangChain
+hosts receive the full structured decision in `ToolMessage.artifact`; the
+model-facing content remains a short summary. The adapter never executes or
+authorizes the selected service.
+
 A public reference instance runs at **https://asm-spec.onrender.com**. It also dogfoods ASM's own publishing convention: `GET /.well-known/asm` serves the library catalog (one re-stampable `generated_at`, per-manifest links), and `GET /manifest/{service_id}` serves each full manifest — ASM is its own first publisher.
 
 ```bash
@@ -374,9 +401,10 @@ Schema: [`schema/asm-v0.3.schema.json`](schema/asm-v0.3.schema.json).
 
 ```text
 schema/                         ASM JSON Schema
+src/asm_protocol/               Canonical Python SDK: selection, cost, version
 library/                        Tool-value library (agent tool selection) + select_demo.py
 manifests/                      75 source-linked manifests
-scorer/                         Python TOPSIS scorer and tests
+scorer/                         Legacy experimental per-unit TOPSIS scorer and tests
 registry/                       MCP registry server exposing ASM tools
 examples/mcp-server-json/       MCP Registry server.json examples
 docs/integrations/              MCP Registry and aggregator integration docs

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/README.md

@@ -25,7 +25,7 @@ It accepts either a standalone ASM manifest or an MCP Registry `server.json`
 with publisher-provided ASM metadata:
 
 ```bash
-python -m pip install "asm-protocol==0.5.2"
+python -m pip install "asm-protocol==0.6.0"
 asm-lint server.json --format markdown --output asm-lint-report.md
 ```
 
@@ -42,7 +42,7 @@ steps:
   - uses: actions/setup-python@v5
     with:
       python-version: "3.12"
-  - uses: YE-YI7/asm-spec/.github/actions/asm-lint@v0.5.2
+  - uses: YE-YI7/asm-spec/.github/actions/asm-lint@v0.6.0
     with:
       path: server.json
       fail-on: invalid
@@ -51,6 +51,8 @@ steps:
 The Action adds the full Markdown report to the job summary. It does not call
 an ASM-hosted API or upload the inspected manifest. See the
 [lint and CI guide](docs/adoption/asm-lint.md) for status semantics.
+For a bounded producer-side change, use the
+[10-minute adoption package](docs/adoption/ten-minute.md).
 
 ## Try it: pick a tool for a task
 
@@ -59,7 +61,13 @@ git clone https://github.com/YE-YI7/asm-spec.git && cd asm-spec
 python library/select_demo.py
 ```
 
-For *"make a study plan and remind me daily"* with a cloud agent on Windows, the selector drops the tools it can't drive (Apple Reminders, Things 3 — local-device only) and the ones it can't call directly (Any.do — Zapier only), then ranks the rest. Ask for a built-in pomodoro and the pick changes to TickTick. Ask to *"edit an image and lay out a poster"* and it picks free, scriptable **Photopea** over paid Photoshop — and filters **Affinity Designer**, which exposes no automation API at all.
+The deterministic core does **not** pretend to understand the task sentence. The
+caller supplies structured facts such as `taxonomy`, `required_functions`,
+platform, reach, and expected workload; `task` remains audit/display text. If
+taxonomy and required functions are both absent, selection returns
+`under_specified` instead of choosing an unrelated globally cheap tool.
+
+For *"make a study plan and remind me daily"* with a cloud agent on Windows, the selector drops the tools it can't drive (Apple Reminders, Things 3 — local-device only) and the ones it can't call directly (Any.do — Zapier only), then ranks the rest from those explicit constraints. Ask for a built-in pomodoro and the pick changes to TickTick. Ask to *"edit an image and lay out a poster"* and it filters **Affinity Designer**, which exposes no automation API at all.
 
 The library it selects over is in [`library/`](library/) — 30 real tools across task management, creative design, research, communication, developer tools, booking, and real-estate data today, each carrying:
 
@@ -89,31 +97,50 @@ ASM is MCP-compatible: publish a standalone `.well-known/asm`, or embed ASM in M
 ASM ships an MCP server so any MCP client (Claude Desktop, Cursor, an agent host) can call the selector as a tool — no schema adoption required:
 
 ```bash
-python3 -m pip install "asm-protocol[mcp]==0.5.2"
+python3 -m pip install "asm-protocol[mcp]==0.6.0"
 asm-selector                     # stdio MCP server (MCP SDK 2.x)
 ```
 
 It exposes three tools: **`select_tool`** (pick a tool for a task and return its risk/approval policy), **`list_library_tools`**, and **`get_tool_manifest`**. Point your client's MCP config at `asm-selector`, or at `python3 /path/to/asm-spec/asm_selector_mcp.py`; the selector reads `library/` (override with `ASM_LIBRARY_DIR`). The Python server uses the stable MCP SDK 2.x line and supports the modern `2026-07-28` protocol era. The same selector is importable directly: `from library_select import select`.
 
+Cost output has an explicit `known`, `partial`, or `unknown` status. Metered
+prices need expected monthly usage; one-time licenses need an amortization
+period; prose-only free tiers remain unknown because their allowance and reset
+rules are not machine-readable. If every eligible candidate does not have a
+known cost in the same currency, the selector returns `needs_cost_facts` rather
+than guessing. A caller may explicitly request the `capability_breadth` fallback;
+it is never applied implicitly.
+
 The same engine is also available as:
 
 ```bash
 # CLI (human or scripted)
 asm select "find and book a refundable flight" --taxonomy tool.booking.travel \
-  --requires flight_search,flight_order_create --json
+  --requires flight_search,flight_order_create --fallback-policy capability_breadth --json
 
 # Hosted HTTP API (stdlib-only; deploy anywhere that runs Python)
 python asm_select_api.py     # POST /select, GET /tools, GET /healthz on :8787
 ```
 
-LangChain / LangGraph builders get the same selector as a drop-in tool (`pip install langchain-core`):
+DeepSeek Harness developer-preview users can install the native
+[`asm_select` tool adapter](integrations/deepseek-harness/README.md). It uses
+the same HTTP contract and returns the current structured decision, but never
+invokes or authorizes the selected service. The frozen v0.1 receipt is available
+only through an explicit legacy compatibility profile. The adapter defaults to
+a local selector so task text is not sent to a hosted endpoint implicitly.
+
+LangChain / LangGraph builders get the same selector as a packaged drop-in tool:
 
 ```python
-import sys; sys.path += ["asm-spec", "asm-spec/integrations/langchain"]
-from asm_tools import ASMToolSelectorTool
+from asm_protocol.integrations.langchain import ASMToolSelectorTool
 agent_tools = [ASMToolSelectorTool()]   # name: asm_tool_selector
 ```
 
+Install it with `python -m pip install "asm-protocol[langchain]==0.6.0"`. LangChain
+hosts receive the full structured decision in `ToolMessage.artifact`; the
+model-facing content remains a short summary. The adapter never executes or
+authorizes the selected service.
+
 A public reference instance runs at **https://asm-spec.onrender.com**. It also dogfoods ASM's own publishing convention: `GET /.well-known/asm` serves the library catalog (one re-stampable `generated_at`, per-manifest links), and `GET /manifest/{service_id}` serves each full manifest — ASM is its own first publisher.
 
 ```bash
@@ -358,9 +385,10 @@ Schema: [`schema/asm-v0.3.schema.json`](schema/asm-v0.3.schema.json).
 
 ```text
 schema/                         ASM JSON Schema
+src/asm_protocol/               Canonical Python SDK: selection, cost, version
 library/                        Tool-value library (agent tool selection) + select_demo.py
 manifests/                      75 source-linked manifests
-scorer/                         Python TOPSIS scorer and tests
+scorer/                         Legacy experimental per-unit TOPSIS scorer and tests
 registry/                       MCP registry server exposing ASM tools
 examples/mcp-server-json/       MCP Registry server.json examples
 docs/integrations/              MCP Registry and aggregator integration docs

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/asm_cli.py

@@ -515,6 +515,7 @@ def cmd_select(args: argparse.Namespace) -> int:
         required_functions=[f for f in (args.requires or "").split(",") if f],
         require_approval_for=[s for s in (args.approval_for or "").split(",") if s],
         require_agent_completable_setup=args.agent_setup_only,
+        fallback_policy=args.fallback_policy,
     )
     if args.json:
         print(json.dumps(result, indent=2, ensure_ascii=False))
@@ -522,18 +523,26 @@ def cmd_select(args: argparse.Namespace) -> int:
 
     sel = result["selected"]
     if not sel:
-        print("No eligible tool.")
+        print(f"No selection ({result['selection_status']}): {result['reason']}")
+        for candidate in result["alternatives"][: args.limit]:
+            print(f"  candidate: {candidate['display_name']} ({candidate['cost_estimate']['status']} cost)")
         for r in result["rejected"][: args.limit]:
             print(f"  - {r['service']}: {r['reason']}")
         return 2
     print(f"Selected: {sel['display_name']} ({sel['service_id']})")
-    print(f"  cost=${sel['monthly_cost_usd']}/mo, interface={sel['interface']}, reach={sel['reach']}")
+    estimate = sel["cost_estimate"]
+    cost = (f"${sel['monthly_cost_usd']}/mo" if sel["monthly_cost_usd"] is not None
+            else f"{estimate['status']} (workload or allowance facts required)")
+    print(f"  cost={cost}, interface={sel['interface']}, reach={sel['reach']}")
     if sel.get("agent_completable_setup") is not None:
         print(f"  setup: agent_completable={sel['agent_completable_setup']}, requires={sel.get('setup_requires', [])}")
     print(f"  risk={result['risk_class']}, approval_required={result['approval_required']}, side_effects={result['side_effects']}")
     print(f"  reason: {result['reason']}")
     for alt in result["alternatives"][: args.limit]:
-        print(f"  alt: {alt['display_name']} (${alt['monthly_cost_usd']}/mo)")
+        alt_cost = (f"${alt['monthly_cost_usd']}/mo"
+                    if alt["monthly_cost_usd"] is not None
+                    else alt["cost_estimate"]["status"])
+        print(f"  alt: {alt['display_name']} ({alt_cost})")
     if result["rejected"]:
         print("  filtered out:")
         for r in result["rejected"][: args.limit]:
@@ -556,6 +565,8 @@ def build_parser() -> argparse.ArgumentParser:
                      help="Comma-separated side-effects that force approval, e.g. financial_charge,sends_message")
     sel.add_argument("--agent-setup-only", action="store_true",
                      help="Drop tools needing human-in-the-loop setup (paid signup, OAuth consent, approval)")
+    sel.add_argument("--fallback-policy", choices=["capability_breadth"],
+                     help="Explicit fallback when eligible candidate costs are not comparable")
     sel.add_argument("--json", action="store_true", help="Emit the structured decision as JSON")
     sel.add_argument("--limit", type=int, default=5, help="Maximum alternatives/rejections to print")
     sel.set_defaults(func=cmd_select)

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/asm_protocol.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: asm-protocol
-Version: 0.5.2
+Version: 0.6.0
 Summary: Agent Service Manifest: value metadata and scoring for autonomous service selection
 Author: Yi Guo
 License-Expression: MIT
@@ -41,7 +41,7 @@ It accepts either a standalone ASM manifest or an MCP Registry `server.json`
 with publisher-provided ASM metadata:
 
 ```bash
-python -m pip install "asm-protocol==0.5.2"
+python -m pip install "asm-protocol==0.6.0"
 asm-lint server.json --format markdown --output asm-lint-report.md
 ```
 
@@ -58,7 +58,7 @@ steps:
   - uses: actions/setup-python@v5
     with:
       python-version: "3.12"
-  - uses: YE-YI7/asm-spec/.github/actions/asm-lint@v0.5.2
+  - uses: YE-YI7/asm-spec/.github/actions/asm-lint@v0.6.0
     with:
       path: server.json
       fail-on: invalid
@@ -67,6 +67,8 @@ steps:
 The Action adds the full Markdown report to the job summary. It does not call
 an ASM-hosted API or upload the inspected manifest. See the
 [lint and CI guide](docs/adoption/asm-lint.md) for status semantics.
+For a bounded producer-side change, use the
+[10-minute adoption package](docs/adoption/ten-minute.md).
 
 ## Try it: pick a tool for a task
 
@@ -75,7 +77,13 @@ git clone https://github.com/YE-YI7/asm-spec.git && cd asm-spec
 python library/select_demo.py
 ```
 
-For *"make a study plan and remind me daily"* with a cloud agent on Windows, the selector drops the tools it can't drive (Apple Reminders, Things 3 — local-device only) and the ones it can't call directly (Any.do — Zapier only), then ranks the rest. Ask for a built-in pomodoro and the pick changes to TickTick. Ask to *"edit an image and lay out a poster"* and it picks free, scriptable **Photopea** over paid Photoshop — and filters **Affinity Designer**, which exposes no automation API at all.
+The deterministic core does **not** pretend to understand the task sentence. The
+caller supplies structured facts such as `taxonomy`, `required_functions`,
+platform, reach, and expected workload; `task` remains audit/display text. If
+taxonomy and required functions are both absent, selection returns
+`under_specified` instead of choosing an unrelated globally cheap tool.
+
+For *"make a study plan and remind me daily"* with a cloud agent on Windows, the selector drops the tools it can't drive (Apple Reminders, Things 3 — local-device only) and the ones it can't call directly (Any.do — Zapier only), then ranks the rest from those explicit constraints. Ask for a built-in pomodoro and the pick changes to TickTick. Ask to *"edit an image and lay out a poster"* and it filters **Affinity Designer**, which exposes no automation API at all.
 
 The library it selects over is in [`library/`](library/) — 30 real tools across task management, creative design, research, communication, developer tools, booking, and real-estate data today, each carrying:
 
@@ -105,31 +113,50 @@ ASM is MCP-compatible: publish a standalone `.well-known/asm`, or embed ASM in M
 ASM ships an MCP server so any MCP client (Claude Desktop, Cursor, an agent host) can call the selector as a tool — no schema adoption required:
 
 ```bash
-python3 -m pip install "asm-protocol[mcp]==0.5.2"
+python3 -m pip install "asm-protocol[mcp]==0.6.0"
 asm-selector                     # stdio MCP server (MCP SDK 2.x)
 ```
 
 It exposes three tools: **`select_tool`** (pick a tool for a task and return its risk/approval policy), **`list_library_tools`**, and **`get_tool_manifest`**. Point your client's MCP config at `asm-selector`, or at `python3 /path/to/asm-spec/asm_selector_mcp.py`; the selector reads `library/` (override with `ASM_LIBRARY_DIR`). The Python server uses the stable MCP SDK 2.x line and supports the modern `2026-07-28` protocol era. The same selector is importable directly: `from library_select import select`.
 
+Cost output has an explicit `known`, `partial`, or `unknown` status. Metered
+prices need expected monthly usage; one-time licenses need an amortization
+period; prose-only free tiers remain unknown because their allowance and reset
+rules are not machine-readable. If every eligible candidate does not have a
+known cost in the same currency, the selector returns `needs_cost_facts` rather
+than guessing. A caller may explicitly request the `capability_breadth` fallback;
+it is never applied implicitly.
+
 The same engine is also available as:
 
 ```bash
 # CLI (human or scripted)
 asm select "find and book a refundable flight" --taxonomy tool.booking.travel \
-  --requires flight_search,flight_order_create --json
+  --requires flight_search,flight_order_create --fallback-policy capability_breadth --json
 
 # Hosted HTTP API (stdlib-only; deploy anywhere that runs Python)
 python asm_select_api.py     # POST /select, GET /tools, GET /healthz on :8787
 ```
 
-LangChain / LangGraph builders get the same selector as a drop-in tool (`pip install langchain-core`):
+DeepSeek Harness developer-preview users can install the native
+[`asm_select` tool adapter](integrations/deepseek-harness/README.md). It uses
+the same HTTP contract and returns the current structured decision, but never
+invokes or authorizes the selected service. The frozen v0.1 receipt is available
+only through an explicit legacy compatibility profile. The adapter defaults to
+a local selector so task text is not sent to a hosted endpoint implicitly.
+
+LangChain / LangGraph builders get the same selector as a packaged drop-in tool:
 
 ```python
-import sys; sys.path += ["asm-spec", "asm-spec/integrations/langchain"]
-from asm_tools import ASMToolSelectorTool
+from asm_protocol.integrations.langchain import ASMToolSelectorTool
 agent_tools = [ASMToolSelectorTool()]   # name: asm_tool_selector
 ```
 
+Install it with `python -m pip install "asm-protocol[langchain]==0.6.0"`. LangChain
+hosts receive the full structured decision in `ToolMessage.artifact`; the
+model-facing content remains a short summary. The adapter never executes or
+authorizes the selected service.
+
 A public reference instance runs at **https://asm-spec.onrender.com**. It also dogfoods ASM's own publishing convention: `GET /.well-known/asm` serves the library catalog (one re-stampable `generated_at`, per-manifest links), and `GET /manifest/{service_id}` serves each full manifest — ASM is its own first publisher.
 
 ```bash
@@ -374,9 +401,10 @@ Schema: [`schema/asm-v0.3.schema.json`](schema/asm-v0.3.schema.json).
 
 ```text
 schema/                         ASM JSON Schema
+src/asm_protocol/               Canonical Python SDK: selection, cost, version
 library/                        Tool-value library (agent tool selection) + select_demo.py
 manifests/                      75 source-linked manifests
-scorer/                         Python TOPSIS scorer and tests
+scorer/                         Legacy experimental per-unit TOPSIS scorer and tests
 registry/                       MCP registry server exposing ASM tools
 examples/mcp-server-json/       MCP Registry server.json examples
 docs/integrations/              MCP Registry and aggregator integration docs

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/asm_protocol.egg-info/SOURCES.txt

@@ -19,10 +19,12 @@ schema/__init__.py
 schema/asm-receipt-envelope-v0.1.schema.json
 schema/asm-v0.2.schema.json
 schema/asm-v0.3.schema.json
+schema/selection-receipt-v0.1.schema.json
 scorer/__init__.py
 scorer/scorer.py
 scorer/test_asm_lint.py
 scorer/test_asm_selector_mcp.py
+scorer/test_cost_estimation.py
 scorer/test_data_quality_audit.py
 scorer/test_langchain_adapter.py
 scorer/test_library_select.py
@@ -30,5 +32,13 @@ scorer/test_manifests_schema.py
 scorer/test_mcp_server_json_asm.py
 scorer/test_openrouter_adapter.py
 scorer/test_scorer.py
+scorer/test_version_contract.py
 scorer/test_x402_bridge.py
-scorer/data/elo_snapshot.json
+scorer/data/__init__.py
+scorer/data/elo_snapshot.json
+src/asm_protocol/__init__.py
+src/asm_protocol/cost.py
+src/asm_protocol/selection.py
+src/asm_protocol/version.py
+src/asm_protocol/integrations/__init__.py
+src/asm_protocol/integrations/langchain.py

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/asm_protocol.egg-info/top_level.txt

@@ -1,6 +1,7 @@
 _asm_library_data
 asm_cli
 asm_lint
+asm_protocol
 asm_schema
 asm_select_api
 asm_selector_mcp

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/asm_select_api.py

@@ -4,7 +4,8 @@
 Endpoints:
   POST /select          body: {task, taxonomy?, agent_reach?, user_platform?,
                                required_functions?, require_approval_for?,
-                               require_agent_completable_setup?}
+                               require_agent_completable_setup?, workload?: {
+                                 monthly_units?, amortization_months?}}
                         -> structured selection decision (same shape as library_select.select)
   GET  /tools           ?taxonomy=...  -> list of selectable tools
   GET  /.well-known/asm -> ASM catalog: one re-stampable generated_at + per-manifest
@@ -29,7 +30,7 @@ from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from pathlib import Path
 from urllib.parse import parse_qs, unquote, urlparse
 
-from library_select import load_library, monthly_cost, select
+from library_select import estimate_monthly_cost, load_library, select
 
 _LIBRARY = load_library()
 _BY_ID = {m.get("service_id"): m for m in _LIBRARY}
@@ -138,6 +139,7 @@ def _tools_listing(taxonomy: str | None) -> list[dict]:
         if taxonomy and m.get("taxonomy") != taxonomy:
             continue
         inv = m.get("invocation") or {}
+        estimate = estimate_monthly_cost(m)
         out.append({
             "service_id": m.get("service_id"),
             "display_name": m.get("display_name"),
@@ -146,7 +148,8 @@ def _tools_listing(taxonomy: str | None) -> list[dict]:
             "reach": inv.get("reach"),
             "agent_operable": inv.get("agent_operable"),
             "agent_completable_setup": inv.get("agent_completable_setup"),
-            "monthly_cost_usd": round(monthly_cost(m), 2),
+            "monthly_cost_usd": estimate.monthly_total,
+            "cost_estimate": estimate.to_dict(),
         })
     return out
 
@@ -248,18 +251,25 @@ class Handler(BaseHTTPRequestHandler):
         if not task:
             self._send(400, {"error": "'task' is required"})
             return
-        result = select(
-            task,
-            taxonomy=req.get("taxonomy"),
-            agent_reach=req.get("agent_reach", "cloud"),
-            user_platform=req.get("user_platform", "any"),
-            required_functions=req.get("required_functions") or [],
-            require_approval_for=req.get("require_approval_for") or [],
-            require_agent_completable_setup=bool(req.get("require_agent_completable_setup", False)),
-            library=_LIBRARY,
-            receipt=bool(req.get("receipt", False)),
-        )
-        self._send(200, result)
+        try:
+            result = select(
+                task,
+                taxonomy=req.get("taxonomy"),
+                agent_reach=req.get("agent_reach", "cloud"),
+                user_platform=req.get("user_platform", "any"),
+                required_functions=req.get("required_functions") or [],
+                require_approval_for=req.get("require_approval_for") or [],
+                require_agent_completable_setup=bool(req.get("require_agent_completable_setup", False)),
+                workload=req.get("workload"),
+                fallback_policy=req.get("fallback_policy"),
+                selection_profile=req.get("selection_profile", "current"),
+                library=_LIBRARY,
+                receipt=bool(req.get("receipt", False)),
+            )
+        except (TypeError, ValueError) as error:
+            self._send(400, {"error": str(error)})
+            return
+        self._send(422 if result["selection_status"] == "under_specified" else 200, result)
 
     def log_message(self, fmt, *args):  # quiet default logging
         pass

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/asm_selector_mcp.py

@@ -9,15 +9,16 @@ policy (risk / approval / side-effects) to gate the action.
 Run (stdio):  python asm_selector_mcp.py     (or `asm-selector` once installed)
 Requires:     pip install "asm-protocol[mcp]"
 """
+
 from __future__ import annotations
 
 from mcp.server import MCPServer
 
-from library_select import load_library, monthly_cost, select
+from library_select import SELECTOR_VERSION, estimate_monthly_cost, load_library, select
 
 mcp = MCPServer(
     "asm-selector",
-    version="0.5.1",
+    version=SELECTOR_VERSION.removeprefix("asm-protocol/"),
     description="Select agent-operable tools using ASM value and policy metadata.",
 )
 
@@ -31,6 +32,9 @@ def select_tool(
     required_functions: list[str] | None = None,
     require_approval_for: list[str] | None = None,
     require_agent_completable_setup: bool = False,
+    monthly_usage: dict[str, float] | None = None,
+    amortization_months: int | None = None,
+    fallback_policy: str | None = None,
 ) -> dict:
     """Pick the best tool from the ASM library for a task.
 
@@ -40,7 +44,7 @@ def select_tool(
     policy so the caller can gate high-stakes actions before invoking.
 
     Args:
-        task: natural-language description of what the agent must do.
+        task: audit/display text; the deterministic core does not interpret it.
         taxonomy: optional ASM taxonomy to scope candidates (e.g. tool.booking.travel).
         agent_reach: 'cloud' (remote/headless) or 'local_device' (runs on user's machine).
         user_platform: e.g. windows, macos, ios, android, web, any.
@@ -48,6 +52,9 @@ def select_tool(
         require_approval_for: side-effects that should force approval (e.g. ["financial_charge","sends_message"]).
         require_agent_completable_setup: if true, drop tools that need a human-in-the-loop
             step (paid signup, OAuth consent, manual approval) the agent cannot complete unattended.
+        monthly_usage: expected monthly units keyed by billing dimension.
+        amortization_months: period for normalizing a declared one-time purchase.
+        fallback_policy: optional 'capability_breadth'; never applied implicitly.
     """
     return select(
         task,
@@ -57,27 +64,35 @@ def select_tool(
         required_functions=required_functions or [],
         require_approval_for=require_approval_for or [],
         require_agent_completable_setup=require_agent_completable_setup,
+        workload={
+            "monthly_units": monthly_usage,
+            "amortization_months": amortization_months,
+        },
+        fallback_policy=fallback_policy,
     )
 
 
 @mcp.tool()
 def list_library_tools(taxonomy: str | None = None) -> list[dict]:
-    """List tools in the ASM library (optionally filtered by taxonomy), with
-    interface/reach and monthly cost so a client can browse what's selectable."""
+    """List tools with invocation facts and honest cost-estimate status."""
     out = []
     for m in load_library():
         if taxonomy and m.get("taxonomy") != taxonomy:
             continue
         inv = m.get("invocation") or {}
-        out.append({
-            "service_id": m.get("service_id"),
-            "display_name": m.get("display_name"),
-            "taxonomy": m.get("taxonomy"),
-            "interface": inv.get("interface"),
-            "reach": inv.get("reach"),
-            "agent_operable": inv.get("agent_operable"),
-            "monthly_cost_usd": round(monthly_cost(m), 2),
-        })
+        estimate = estimate_monthly_cost(m)
+        out.append(
+            {
+                "service_id": m.get("service_id"),
+                "display_name": m.get("display_name"),
+                "taxonomy": m.get("taxonomy"),
+                "interface": inv.get("interface"),
+                "reach": inv.get("reach"),
+                "agent_operable": inv.get("agent_operable"),
+                "monthly_cost_usd": estimate.monthly_total,
+                "cost_estimate": estimate.to_dict(),
+            }
+        )
     return out
 
 

asm_protocol-0.6.0/library_select.py

@@ -0,0 +1,11 @@
+#!/usr/bin/env python3
+"""Backward-compatible facade for the canonical :mod:`asm_protocol` SDK."""
+
+import sys
+from pathlib import Path
+
+_SRC = Path(__file__).resolve().parent / "src"
+if str(_SRC) not in sys.path:
+    sys.path.insert(0, str(_SRC))
+
+from asm_protocol.selection import *

{asm_protocol-0.5.2 → asm_protocol-0.6.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "asm-protocol"
-version = "0.5.2"
+dynamic = ["version"]
 description = "Agent Service Manifest: value metadata and scoring for autonomous service selection"
 readme = "README.md"
 requires-python = ">=3.10"
@@ -27,11 +27,21 @@ asm-selector = "asm_selector_mcp:main"
 
 [tool.setuptools]
 py-modules = ["asm_cli", "asm_lint", "mcp_server_json_asm", "openrouter_adapter", "library_select", "asm_selector_mcp", "asm_select_api", "_asm_library_data"]
-packages = ["scorer", "asm_schema"]
+packages = [
+  "asm_protocol",
+  "asm_protocol.integrations",
+  "scorer",
+  "scorer.data",
+  "asm_schema",
+]
 
 [tool.setuptools.package-dir]
+asm_protocol = "src/asm_protocol"
 asm_schema = "schema"
 
+[tool.setuptools.dynamic]
+version = {attr = "asm_protocol.version.__version__"}
+
 [tool.setuptools.package-data]
 scorer = ["data/*.json"]
 asm_schema = ["*.json"]
@@ -39,3 +49,4 @@ asm_schema = ["*.json"]
 [tool.pytest.ini_options]
 testpaths = ["scorer", "tools/asm-gen"]
 python_files = ["test_*.py"]
+pythonpath = [".", "src"]