feat: establish the Target architecture as a graph before code

Design is now approved before any infrastructure code is written (ADR-0005).
A new target-architect Specialist produces one graph of Elements,
Relationships and Checks; the diagram is rendered from it, deployment
guidance must account for every Check, and validation gives a verdict per
Check. Requirements, components, products and recommendations carry ids in
data files, and a submission with a gap in the Requirements chain is refused.

Also in this commit:
- Customer cases group Executions, with an intent history per case
- lesson candidates are captured when an Execution ends
- evaluation gains chain, verdict, judged and interactive checks, and
  per-case thresholds
- Terraform providers are cached between checks
- the requirements analyst proposes defaults instead of asking
- the legacy-baseline case and the checks that only read it are removed
- ADR-0006 (Change requests) accepted; ADR-0004 and ADR-0007 proposed
- plans for iterations 4 to 7 under .agents/plans/

Unproven with a real model: no evaluation run has completed on this design.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-10-07 13:59:09 +01:00
parent 7ee49ab352
commit e9246828b3
586 changed files with 5166 additions and 22432 deletions

View File

@@ -56,6 +56,7 @@ class SolutionArchitectExecutor(AgentExecutor):
interaction = await self._engine.start(
context.get_user_input(),
session_id=context.context_id,
customer_case=options.get("customer_case"),
execution_id=task_id,
pre_approved=bool(options.get("pre_approved", False)),
allow_cloud_read=options.get("allow_cloud_read"),

View File

@@ -31,6 +31,7 @@ from google.adk.tools.tool_context import ToolContext
from google.genai import types
from app.adk.runtime import (
REJECTION_PREFIXES,
AgentTask,
BudgetExhausted,
Outcome,
@@ -52,13 +53,6 @@ TIMED_OUT = (
"Your last model request timed out, so its reply was lost. Continue the task, "
"and finish by calling the tool your instructions name for ending your turn."
)
# Tool replies that mean the call did nothing, logged so stuck agents can be diagnosed.
REJECTION_PREFIXES = (
"Not submitted",
"Refused",
"Discovery gap",
"terraform validate failed",
)
# Client-side timeouts from either HTTP stack google-genai may use (httpx or aiohttp,
# whose timeout errors subclass TimeoutError).
MODEL_TIMEOUTS = (TimeoutError, httpx.TimeoutException)

View File

@@ -7,6 +7,7 @@ enforce is exercised exactly as a model would meet it.
from __future__ import annotations
import json
from collections import deque
from collections.abc import Callable, Mapping, Sequence
from dataclasses import dataclass, field
@@ -20,6 +21,7 @@ from app.adk.runtime import (
Stopped,
ToolSpec,
)
from app.workflows.deliverables import DEPENDENCY_GRAPH, DeliverableKind
@dataclass(frozen=True)
@@ -31,6 +33,97 @@ class Call:
Turn = Sequence[Call] | Callable[[AgentTask], Sequence[Call]]
# One small, coherent Requirements chain, so sample submissions pass every link.
SAMPLE_DATA: dict[str, str] = {
"requirements.json": json.dumps(
[
{"id": "FR-1", "kind": "functional", "text": "Accept events."},
{"id": "NFR-1", "kind": "non_functional", "text": "99.9% availability."},
]
),
"components.json": json.dumps(
[{"id": "C-1", "name": "Ingress", "requirements": ["FR-1", "NFR-1"]}]
),
"products.json": json.dumps([{"component": "C-1", "product": "Example Run"}]),
"recommendations.json": json.dumps(
[
{
"id": "DR-1",
"pillar": "security",
"text": "Use a dedicated identity.",
"components": ["C-1"],
}
]
),
"target-architecture.json": json.dumps(
{
"elements": [
{
"id": "E-1",
"name": "Ingress service",
"component": "C-1",
"product": "Example Run",
"resource_type": "example_run_service",
},
{
"id": "E-2",
"name": "Ingress identity",
"product": "Example IAM",
"resource_type": "example_service_account",
},
],
"relationships": [
{"id": "R-1", "from": "E-1", "to": "E-2", "kind": "runs as"}
],
"checks": [
{
"id": "K-1",
"applies_to": "E-1",
"statement": "The service runs at least two instances.",
"source": "NFR-1",
},
{
"id": "K-2",
"applies_to": "E-2",
"statement": "The identity is used by this service only.",
"source": "DR-1",
},
],
}
),
"implementation.json": json.dumps(
[
{"check": "K-1", "file": "terraform/main.tf", "address": "example.run"},
{"check": "K-2", "file": "terraform/main.tf", "address": "example.sa"},
]
),
"verdicts.json": json.dumps(
{
"dry_run": {"verdict": "pass", "evidence": "terraform validate passed."},
"checks": [
{"check": "K-1", "verdict": "pass", "evidence": "min instances is 2."},
{"check": "K-2", "verdict": "pass", "evidence": "One binding."},
],
"proposed_additions": [],
}
),
}
def sample_submission(kind: DeliverableKind, note: str = "") -> Call:
"""A `submit_deliverable` call that meets the Deliverable's contract."""
spec = DEPENDENCY_GRAPH[kind]
parts = [f"# {kind}", note or "Summary."]
parts += [f"## {section}\n{note or 'Details.'}" for section in spec.sections]
attachments = {
name: SAMPLE_DATA.get(name, f"# {name}\n") for name in spec.attachments
}
return Call(
"submit_deliverable",
{"content": "\n".join(parts) + "\n", "attachments": attachments},
)
@dataclass(frozen=True)
class RunRecord:
role: str

View File

@@ -6,7 +6,17 @@ a RoleSpec, an AgentTask and ToolSpecs in; an Outcome back. No framework types.
from __future__ import annotations
from collections.abc import Awaitable, Callable, Mapping, Sequence
import asyncio
import contextvars
from collections.abc import (
AsyncIterator,
Awaitable,
Callable,
Mapping,
MutableMapping,
Sequence,
)
from contextlib import asynccontextmanager
from dataclasses import dataclass, field
from typing import Any, Protocol
@@ -47,10 +57,15 @@ class Submitted:
@dataclass(frozen=True)
class NeedsInput:
"""The agent can't continue without the user. `permission` names a tool tier to grant."""
"""The agent can't continue without the user. `permission` names a tool tier to grant.
`draft` is work the agent wants back when it is started again with the answers.
"""
questions: tuple[str, ...]
permission: str | None = None
draft: str | None = None
draft_attachments: Mapping[str, str] = field(default_factory=dict)
@dataclass(frozen=True)
@@ -127,8 +142,62 @@ class ToolSpec:
tier: str = "general" # general | cloud_read | execution
# The tools a role gets beyond the Execution tools, given the Deliverable it is producing.
ToolFactory = Callable[[RoleSpec, str | None], Sequence[ToolSpec]]
# The tools a role gets beyond the Execution tools, given the Deliverable it is producing
# and that Deliverable's working memory: JSON-compatible data a tool may read and write,
# kept with the Execution across attempts.
ToolFactory = Callable[
[RoleSpec, str | None, MutableMapping[str, Any]], Sequence[ToolSpec]
]
class RunLimiter:
"""Caps how many agent runs are in flight at once.
A run that fans out to further runs lends its slot while it waits, so the cap
holds across the whole hierarchy and a cap of 1 cannot deadlock.
"""
def __init__(self, limit: int) -> None:
if limit < 1:
raise ValueError("limit must be at least 1")
self.limit = limit
self._semaphore = asyncio.Semaphore(limit)
self._holding: contextvars.ContextVar[bool] = contextvars.ContextVar(
"run_limiter_holding", default=False
)
@asynccontextmanager
async def slot(self) -> AsyncIterator[None]:
async with self._semaphore:
token = self._holding.set(True)
try:
yield
finally:
self._holding.reset(token)
@asynccontextmanager
async def lend(self) -> AsyncIterator[None]:
"""Give up the caller's slot, if it holds one, while its own fan-out runs."""
if not self._holding.get():
yield
return
self._semaphore.release()
token = self._holding.set(False)
try:
yield
finally:
self._holding.reset(token)
await self._semaphore.acquire()
# Tool replies that mean the call did nothing. Runtimes log them, and the Execution
# keeps them as lesson candidates.
REJECTION_PREFIXES = (
"Not submitted",
"Refused",
"Discovery gap",
"terraform validate failed",
)
class AgentRuntime(Protocol):

View File

@@ -22,11 +22,14 @@ from langgraph.checkpoint.base import BaseCheckpointSaver
from langgraph.types import Command
from app.adk.roles import load_roles
from app.adk.runtime import AgentRuntime, RoleSpec, ToolFactory
from app.adk.runtime import AgentRuntime, RoleSpec, RunLimiter, ToolFactory
from app.checkpointing import open_checkpointer
from app.config import Settings
from app.customer_case import record_intent
from app.providers import ProviderBundle, load_provider
from app.recorder import DatabaseRecorder, ExecutionRecorder
from app.skills.agent_skills import skill_version
from app.states.execution_state import check_customer_case
from app.states.ledger import Ledger
from app.tools.execution import ORCHESTRATOR
from app.tools.local import TerraformRunner, local_tools
@@ -62,6 +65,8 @@ class ExecutionEngine:
max_revisions: int = 3,
max_orchestrator_turns: int = 100,
tool_factory: ToolFactory | None = None,
limiter: RunLimiter | None = None,
skill_versions: Mapping[str, str] | None = None,
) -> None:
required = {
ORCHESTRATOR,
@@ -71,6 +76,7 @@ class ExecutionEngine:
if missing:
raise ValueError(f"Missing roles: {', '.join(missing)}")
self._max_revisions = max_revisions
self._output_root = output_root
self._graph = build_execution_graph(
runtime=runtime,
roles=roles,
@@ -78,6 +84,8 @@ class ExecutionEngine:
output_root=output_root,
max_orchestrator_turns=max_orchestrator_turns,
tool_factory=tool_factory,
limiter=limiter,
skill_versions=skill_versions,
).compile(checkpointer=checkpointer)
async def start(
@@ -85,6 +93,7 @@ class ExecutionEngine:
workflow_request: str,
*,
session_id: str | None = None,
customer_case: str | None = None,
pre_approved: bool = False,
allow_cloud_read: bool | None = None,
execution_id: str | None = None,
@@ -92,14 +101,25 @@ class ExecutionEngine:
) -> Interaction:
"""Start an Execution. `scripted_answers` are (regex, answer) pairs a
pre-approved Execution uses to answer questions, for evaluation runs."""
customer_case = check_customer_case(customer_case)
execution_id = execution_id or str(uuid.uuid4())
if await self._exists(execution_id):
raise ValueError(f"Execution {execution_id} already exists")
if pre_approved and allow_cloud_read is None:
allow_cloud_read = False
session_id = session_id or str(uuid.uuid4())
intent_version = record_intent(
self._output_root,
customer_case,
execution_id=execution_id,
session_id=session_id,
workflow_request=workflow_request,
)
initial: dict[str, Any] = {
"execution_id": execution_id,
"session_id": session_id or str(uuid.uuid4()),
"session_id": session_id,
"customer_case": customer_case,
"intent_version": intent_version,
"workflow_request": workflow_request,
"pre_approved": pre_approved,
"allow_cloud_read": allow_cloud_read,
@@ -112,9 +132,12 @@ class ExecutionEngine:
"pending_questions": [],
"awaiting_questions": [],
"answers": {},
"drafts": {},
"working": {},
"pending_permission": None,
"prompt": None,
"outcomes": [],
"refusals": [],
"outcomes_applied": 0,
"events_recorded": 0,
"turns": 0,
@@ -218,15 +241,18 @@ async def open_assembly(
runner = TerraformRunner(
binary=settings.TERRAFORM_BINARY,
timeout_seconds=settings.TERRAFORM_TIMEOUT_SECONDS,
plugin_cache_dir=settings.TF_PLUGIN_CACHE_DIR,
plugin_cache_dir=settings.terraform_plugin_cache(),
)
offered = [name for names in bundle.tools.values() for name in names]
# One cap for Specialists and the pillar advisors they fan out to.
limiter = RunLimiter(settings.MAX_PARALLEL_SPECIALISTS)
tool_factory = make_tool_factory(
runtime,
roles,
bundle.pillar_skills,
local_tools=local_tools(offered, runner),
catalog=catalog,
limiter=limiter,
)
async with open_checkpointer(settings) as checkpointer:
@@ -239,5 +265,9 @@ async def open_assembly(
max_revisions=settings.MAX_REVISIONS,
max_orchestrator_turns=settings.MAX_ORCHESTRATOR_TURNS,
tool_factory=tool_factory,
limiter=limiter,
skill_versions={
name: skill_version(meta.path) for name, meta in bundle.skills.items()
},
)
yield Assembly(engine, bundle, roles, dict(catalog.gaps))

View File

@@ -38,7 +38,9 @@ class Settings(BaseSettings):
# Local Terraform tools (fmt, validate, plan; never apply)
TERRAFORM_BINARY: str = "terraform"
TERRAFORM_TIMEOUT_SECONDS: float = 300
TF_PLUGIN_CACHE_DIR: str | None = None
# Where Terraform providers are kept between checks, so they download once.
# Relative paths are under the repository root. Set it empty to disable the cache.
TF_PLUGIN_CACHE_DIR: str = ".cache/terraform-plugins"
# Models: a request with no response in this time is abandoned and retried
MODEL_TIMEOUT_SECONDS: float = 180
@@ -51,12 +53,21 @@ class Settings(BaseSettings):
SPECIALIST_BUDGET: int = 15
MAX_REVISIONS: int = 3
MAX_ORCHESTRATOR_TURNS: int = 100
# Agent runs in flight at once, pillar advisors included (model rate limits)
MAX_PARALLEL_SPECIALISTS: int = 4
CHECKPOINTER: str = "sqlite" # memory | sqlite | postgres
CHECKPOINT_DB_PATH: str = "checkpoints.db"
# Paths
BASE_DIR: Path = Path(__file__).resolve().parent.parent
def terraform_plugin_cache(self) -> str | None:
"""The provider cache directory as an absolute path; None when disabled."""
if not self.TF_PLUGIN_CACHE_DIR.strip():
return None
path = Path(self.TF_PLUGIN_CACHE_DIR).expanduser()
return str(path if path.is_absolute() else self.BASE_DIR / path)
def get_database_url(self) -> str:
"""Construct database connection URL (PostgreSQL or SQLite fallback)."""
if self.DATABASE_URL:

105
app/customer_case.py Normal file
View File

@@ -0,0 +1,105 @@
"""A Customer case on disk: its intent and the history of that intent.
deliverables/cases/<customer_case>/
intent/
001.md, 002.md, ... each statement of what the customer wants, as given
history.json every statement in order, the Execution it started
and how that Execution ended
executions/<execution_id>/ what was generated for one statement
The intent is recorded when an Execution starts, so a request is on record even if
nothing was delivered for it. history.json is the index of the case.
"""
from __future__ import annotations
import json
from datetime import UTC, datetime
from pathlib import Path
from typing import Any
from app.states.execution_state import case_dir
INTENT_DIR = "intent"
HISTORY_FILENAME = "history.json"
IN_PROGRESS = "in_progress"
def intent_dir(output_root: Path, customer_case: str) -> Path:
return case_dir(output_root, customer_case) / INTENT_DIR
def read_history(output_root: Path, customer_case: str) -> dict[str, Any]:
"""The case's intent history; empty when the case has none yet."""
path = intent_dir(output_root, customer_case) / HISTORY_FILENAME
if path.is_file():
try:
history = json.loads(path.read_text(encoding="utf-8"))
if isinstance(history, dict) and isinstance(history.get("intents"), list):
return history
except ValueError:
pass
return {"customer_case": customer_case, "intents": []}
def record_intent(
output_root: Path,
customer_case: str,
*,
execution_id: str,
session_id: str,
workflow_request: str,
) -> int:
"""Record a new statement of the customer's intent and return its version."""
folder = intent_dir(output_root, customer_case)
folder.mkdir(parents=True, exist_ok=True)
history = read_history(output_root, customer_case)
version = max((item["version"] for item in history["intents"]), default=0) + 1
recorded_at = datetime.now(UTC).isoformat(timespec="seconds")
statement = f"{version:03d}.md"
(folder / statement).write_text(
f"# Intent {version}: {customer_case}\n\n"
f"- Recorded: {recorded_at}\n"
f"- Execution: `{execution_id}`\n\n"
f"## What the customer wants\n\n{workflow_request.strip()}\n",
encoding="utf-8",
)
history["intents"].append(
{
"version": version,
"recorded_at": recorded_at,
"statement": f"{INTENT_DIR}/{statement}",
"workflow_request": workflow_request,
"execution_id": execution_id,
"session_id": session_id,
"status": IN_PROGRESS,
"execution": None,
}
)
_write_history(folder, history)
return version
def close_intent(
output_root: Path,
customer_case: str,
*,
execution_id: str,
status: str,
has_files: bool,
) -> None:
"""Record how the Execution started by an intent ended, and where its files are."""
history = read_history(output_root, customer_case)
for item in history["intents"]:
if item.get("execution_id") == execution_id:
item["status"] = status
item["execution"] = f"executions/{execution_id}" if has_files else None
_write_history(intent_dir(output_root, customer_case), history)
return
def _write_history(folder: Path, history: dict[str, Any]) -> None:
folder.mkdir(parents=True, exist_ok=True)
(folder / HISTORY_FILENAME).write_text(
json.dumps(history, indent=2, ensure_ascii=False) + "\n", encoding="utf-8"
)

View File

@@ -68,9 +68,12 @@ def create_app(
interaction = await assembly.engine.start(
workflow_request,
session_id=body.get("session_id"),
customer_case=body.get("customer_case"),
pre_approved=True,
allow_cloud_read=bool(body.get("allow_cloud_read", False)),
)
except ValueError as exc:
return JSONResponse({"error": str(exc)}, status_code=400)
except Exception as exc:
logger.exception("/generate failed")
return JSONResponse(

View File

@@ -6,10 +6,12 @@ append outcomes that the next Orchestrator turn applies in order.
from __future__ import annotations
import dataclasses
import json
import logging
import re
import time
from collections.abc import Awaitable, Callable, Mapping, Sequence
from collections.abc import Awaitable, Callable, Mapping, MutableMapping, Sequence
from dataclasses import dataclass
from pathlib import Path
from typing import Any
@@ -17,22 +19,32 @@ from typing import Any
from langgraph.types import Command, Send, interrupt
from app.adk.runtime import (
REJECTION_PREFIXES,
AgentRuntime,
AgentTask,
BudgetExhausted,
NeedsInput,
Outcome,
RoleSpec,
RunLimiter,
Stopped,
Submitted,
ToolFactory,
ToolSpec,
TurnEnded,
)
from app.customer_case import close_intent
from app.recorder import ExecutionRecorder
from app.states.execution_state import ExecutionState
from app.states.execution_state import (
UNASSIGNED_CASE,
ExecutionState,
execution_dir,
lessons_file,
)
from app.states.ledger import Ledger, LedgerRefusal
from app.states.lessons import collect_lesson_candidates, refusal
from app.tools.execution import ORCHESTRATOR, OrchestratorToolbox, specialist_tools
from app.workflows.chain import VERDICTS_FILE, validation_outcome
from app.workflows.deliverables import (
DEPENDENCY_GRAPH,
PHASE_ORDER,
@@ -47,9 +59,12 @@ NO_USER_REPLY = (
"'Assumptions', and proceed."
)
GUIDE_FILENAME = "solution-architecture-guide.md"
MANIFEST_FILENAME = "execution.json" # request, status and files of one Execution
def _no_tools(role: RoleSpec, deliverable: str | None) -> Sequence[ToolSpec]:
def _no_tools(
role: RoleSpec, deliverable: str | None, working: MutableMapping[str, Any]
) -> Sequence[ToolSpec]:
return ()
@@ -73,9 +88,42 @@ def make_execution_nodes(
output_root: Path,
max_orchestrator_turns: int = 100,
tool_factory: ToolFactory | None = None,
limiter: RunLimiter | None = None,
skill_versions: Mapping[str, str] | None = None,
) -> ExecutionNodes:
tool_factory = tool_factory or _no_tools
def close(state: ExecutionState, status: str, *, has_files: bool) -> None:
close_intent(
output_root,
_case(state),
execution_id=state["execution_id"],
status=status,
has_files=has_files,
)
async def record_lessons(
state: ExecutionState, ledger: Ledger, status: str
) -> None:
record = collect_lesson_candidates(
execution_id=state["execution_id"],
customer_case=_case(state),
status=status,
ledger=ledger,
outcomes=state.get("outcomes", []),
orchestrator_refusals=state.get("refusals", []),
role_skills={name: role.skills for name, role in roles.items()},
skill_versions=skill_versions or {},
)
# Kept apart from deliverables/: an Execution that ends with warnings writes
# no Deliverable files, but what went wrong in it is still worth keeping.
path = lessons_file(output_root, state["execution_id"], _case(state))
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(
json.dumps(record, indent=2, ensure_ascii=False), encoding="utf-8"
)
await recorder.record_lesson_candidates(state["execution_id"], record)
async def record_progress(
state: ExecutionState, ledger: Ledger, status: str | None = None
) -> int:
@@ -102,6 +150,7 @@ def make_execution_nodes(
ledger.revisions(kind) for kind in DEPENDENCY_GRAPH
),
"warnings": state.get("warnings", []),
"customer_case": _case(state),
},
)
return len(ledger.events)
@@ -113,6 +162,7 @@ def make_execution_nodes(
notes = list(state.get("notes", []))
pending_permission = state.get("pending_permission")
warnings = list(state.get("warnings", []))
drafts = dict(state.get("drafts", {}))
outcomes = state.get("outcomes", [])
for record in outcomes[state.get("outcomes_applied", 0) :]:
@@ -124,6 +174,7 @@ def make_execution_nodes(
"content": record["content"],
"attachments": record["attachments"],
}
drafts.pop(kind.value, None)
await recorder.record_submission(
state["execution_id"],
kind.value,
@@ -143,6 +194,11 @@ def make_execution_nodes(
"requested_by": [*requested, kind.value],
}
continue
if record.get("draft"):
drafts[kind.value] = {
"content": record["draft"],
"attachments": record.get("draft_attachments", {}),
}
questions = record.get("questions") or [
(
f"The {record['role']} stopped: {record['summary']} "
@@ -154,6 +210,7 @@ def make_execution_nodes(
turns = state.get("turns", 0) + 1
update: dict[str, Any] = {
"deliverables": deliverables,
"drafts": drafts,
"pending_questions": pending_questions,
"pending_permission": pending_permission,
"outcomes_applied": len(outcomes),
@@ -180,13 +237,21 @@ def make_execution_nodes(
return finish_with("end_with_warnings", warnings=warnings)
toolbox = OrchestratorToolbox(ledger, deliverables, pending_questions)
refusals = list(state.get("refusals", []))
update["refusals"] = refusals
task = AgentTask(
role=ORCHESTRATOR,
prompt="Decide the next step for this Execution using your tools.",
context=_orchestrator_context(state, ledger, pending_questions, notes),
context=_orchestrator_context(
state, ledger, deliverables, pending_questions, notes
),
)
started = time.monotonic()
outcome = await runtime.run(roles[ORCHESTRATOR], task, toolbox.tools())
outcome = await runtime.run(
roles[ORCHESTRATOR],
task,
_noting_refusals(toolbox.tools(), refusals, ORCHESTRATOR, None),
)
logger.info(
"[%s] orchestrator turn %d (%s phase) ended in %.0fs: %s",
state["execution_id"],
@@ -209,7 +274,9 @@ def make_execution_nodes(
sends = [
Send(
"specialist_task",
_specialist_input(state, ledger, deliverables, kind, attempt),
_specialist_input(
state, ledger, deliverables, drafts, kind, attempt
),
)
for kind, attempt in toolbox.started.items()
]
@@ -257,10 +324,18 @@ def make_execution_nodes(
async def specialist_task(task_input: dict[str, Any]) -> dict[str, Any]:
role = roles[task_input["role"]]
kind = DeliverableKind(task_input["kind"])
tools = specialist_tools(
kind,
extra=tool_factory(role, kind.value),
allow_cloud_read=task_input["allow_cloud_read"],
working: dict[str, Any] = dict(task_input.get("working") or {})
refusals: list[dict[str, Any]] = []
tools = _noting_refusals(
specialist_tools(
kind,
extra=tool_factory(role, kind.value, working),
allow_cloud_read=task_input["allow_cloud_read"],
upstream=task_input["context"].get("dependencies"),
),
refusals,
role.name,
kind.value,
)
task = AgentTask(
role=role.name,
@@ -277,7 +352,11 @@ def make_execution_nodes(
)
started = time.monotonic()
try:
outcome: Outcome = await runtime.run(role, task, tools)
if limiter:
async with limiter.slot():
outcome: Outcome = await runtime.run(role, task, tools)
else:
outcome = await runtime.run(role, task, tools)
except Exception as exc:
logger.exception("Specialist %s failed on %s", role.name, kind)
outcome = Stopped(f"it failed with {type(exc).__name__}: {exc}")
@@ -291,14 +370,18 @@ def make_execution_nodes(
)
return {
"outcomes": [
_outcome_record(role.name, kind, task_input["attempt"], outcome)
]
{
**_outcome_record(role.name, kind, task_input["attempt"], outcome),
"refusals": refusals,
}
],
"working": {kind.value: working},
}
async def await_user(state: ExecutionState) -> Command:
prompt = state["prompt"] or {}
if state.get("pre_approved"):
reply = _automatic_reply(prompt, state.get("scripted_answers", []))
reply = automatic_reply(prompt, state.get("scripted_answers", []))
else:
reply = interrupt(prompt)
@@ -359,16 +442,17 @@ def make_execution_nodes(
async def finalize(state: ExecutionState) -> dict[str, Any]:
ledger = Ledger.from_dict(state["ledger"])
_write_outputs(
output_root / "deliverables" / "executions" / state["execution_id"],
state.get("deliverables", {}),
)
_write_outputs(output_root, state)
close(state, "completed", has_files=True)
await record_progress(state, ledger, "completed")
await record_lessons(state, ledger, "completed")
return {"status": "completed"}
async def end_with_warnings(state: ExecutionState) -> dict[str, Any]:
ledger = Ledger.from_dict(state["ledger"])
close(state, "completed_with_warnings", has_files=False)
await record_progress(state, ledger, "completed_with_warnings")
await record_lessons(state, ledger, "completed_with_warnings")
return {"status": "completed_with_warnings"}
return ExecutionNodes(
@@ -383,6 +467,28 @@ def make_execution_nodes(
# --- helpers ---------------------------------------------------------------------------
def _noting_refusals(
tools: Sequence[ToolSpec],
sink: list[dict[str, Any]],
role: str,
deliverable: str | None,
) -> list[ToolSpec]:
"""The same tools, noting each reply that means the call did nothing."""
def noting(tool: ToolSpec) -> ToolSpec:
async def handler(**kwargs: Any) -> Any:
result = await tool.handler(**kwargs)
if result.outcome is None and result.message.startswith(REJECTION_PREFIXES):
sink.append(
refusal(role, deliverable, tool.name, tool.tier, result.message)
)
return result
return dataclasses.replace(tool, handler=handler)
return [noting(tool) for tool in tools]
def _summarise_outcome(outcome: Outcome) -> str:
if isinstance(outcome, TurnEnded):
return f"called {outcome.tool}"
@@ -405,10 +511,19 @@ def _current_phase(ledger: Ledger) -> str:
def _orchestrator_context(
state: ExecutionState,
ledger: Ledger,
deliverables: Mapping[str, Mapping[str, Any]],
pending_questions: Sequence[Mapping[str, Any]],
notes: Sequence[str],
) -> dict[str, Any]:
validation = deliverables.get(DeliverableKind.VALIDATION_RESULTS.value)
return {
# The verdict on each Check, once validation is submitted: failed Checks are
# faults in the code; the rest are gaps in the Target architecture.
"validation": validation_outcome(
validation.get("attachments", {}).get(VERDICTS_FILE, "")
)
if validation
else None,
"workflow_request": state["workflow_request"],
"pre_approved": bool(state.get("pre_approved")),
"allow_cloud_read": state.get("allow_cloud_read"),
@@ -438,6 +553,7 @@ def _specialist_input(
state: ExecutionState,
ledger: Ledger,
deliverables: Mapping[str, Mapping[str, Any]],
drafts: Mapping[str, Mapping[str, Any]],
kind: DeliverableKind,
attempt: int,
) -> dict[str, Any]:
@@ -448,15 +564,18 @@ def _specialist_input(
"kind": kind.value,
"attempt": attempt,
"allow_cloud_read": state.get("allow_cloud_read"),
"working": dict(state.get("working", {}).get(kind.value, {})),
"context": {
"workflow_request": state["workflow_request"],
"pre_approved": bool(state.get("pre_approved")),
"contract": spec.contract(),
"dependencies": {
dependency.value: dict(deliverables[dependency.value])
for dependency in inputs_of(kind)
},
"feedback": list(ledger.feedback(kind)),
"answers": list(state.get("answers", {}).get(kind.value, [])),
"draft": dict(drafts[kind.value]) if kind.value in drafts else None,
"attempt": attempt,
},
}
@@ -479,6 +598,8 @@ def _outcome_record(
"type": "needs_input",
"questions": list(outcome.questions),
"permission": outcome.permission,
"draft": outcome.draft,
"draft_attachments": dict(outcome.draft_attachments),
}
if isinstance(outcome, BudgetExhausted):
return {**base, "type": "budget_exhausted", "summary": outcome.summary}
@@ -495,7 +616,7 @@ def _describe_unfinished(outcome: Outcome) -> str:
return f"Ended with an unexpected outcome: {type(outcome).__name__}."
def _automatic_reply(
def automatic_reply(
prompt: Mapping[str, Any], scripted_answers: Sequence[Sequence[str]] = ()
) -> dict[str, Any]:
"""The reply in a pre-approved Execution: approve, deny cloud reads, and answer
@@ -543,18 +664,44 @@ def _render_prompt(prompt: Mapping[str, Any]) -> str:
return "\n".join(item["question"] for item in prompt.get("questions", []))
def _write_outputs(
out_dir: Path, deliverables: Mapping[str, Mapping[str, Any]]
) -> None:
out_dir.mkdir(parents=True, exist_ok=True)
def _case(state: ExecutionState) -> str:
return state.get("customer_case") or UNASSIGNED_CASE
def _write_outputs(output_root: Path, state: ExecutionState) -> None:
"""Write a finished Execution's files under its Customer case, with a manifest
that ties the request to what was generated for it."""
customer_case = _case(state)
execution_id = state["execution_id"]
out_dir = execution_dir(output_root, execution_id, customer_case)
deliverables: Mapping[str, Mapping[str, Any]] = state.get("deliverables", {})
files: dict[str, str] = {}
by_kind: dict[str, dict[str, Any]] = {}
guide = deliverables.get(DeliverableKind.SOLUTION_GUIDE.value)
if guide:
(out_dir / GUIDE_FILENAME).write_text(guide["content"], encoding="utf-8")
files[GUIDE_FILENAME] = guide["content"]
for kind, deliverable in deliverables.items():
path = out_dir / "deliverables" / f"{kind}.md"
document = f"deliverables/{kind}.md"
files[document] = deliverable["content"]
attachments = dict(deliverable.get("attachments", {}))
files.update(attachments)
by_kind[kind] = {"document": document, "attachments": sorted(attachments)}
for name, content in files.items():
path = out_dir / name
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(deliverable["content"], encoding="utf-8")
for name, content in deliverable.get("attachments", {}).items():
attachment = out_dir / name
attachment.parent.mkdir(parents=True, exist_ok=True)
attachment.write_text(content, encoding="utf-8")
path.write_text(content, encoding="utf-8")
manifest = {
"customer_case": customer_case,
"execution_id": execution_id,
"session_id": state.get("session_id"),
"intent_version": state.get("intent_version"),
"workflow_request": state["workflow_request"],
"status": "completed",
"solution_guide": GUIDE_FILENAME if guide else None,
"deliverables": by_kind,
}
out_dir.mkdir(parents=True, exist_ok=True)
(out_dir / MANIFEST_FILENAME).write_text(
json.dumps(manifest, indent=2, ensure_ascii=False), encoding="utf-8"
)

View File

@@ -27,6 +27,7 @@ def describe_prompt(prompt: Mapping[str, Any]) -> str:
return (
f"Please review the {approval['phase']} Phase: {approval['summary']}\n\n"
f"Deliverables: {', '.join(approval['deliverables'])}.\n\n"
f"{_describe_target_architecture(approval.get('target_architecture'))}"
f"To approve, reply with the data {json.dumps(reply)}. "
f"To request changes, reply with {json.dumps(changes)}. "
"A text reply is recorded as conversation and never approves."
@@ -46,6 +47,24 @@ def describe_prompt(prompt: Mapping[str, Any]) -> str:
return "\n".join(lines) + "\n\nReply in text; answer every question you can."
def _describe_target_architecture(summary: Mapping[str, Any] | None) -> str:
"""What the design Approval commits to: no code is written until it is granted."""
if not summary:
return ""
text = (
f"The Target architecture has {summary['elements']} Elements, "
f"{summary['relationships']} Relationships and {summary['checks']} Checks. "
"No infrastructure code is written until you approve it; the code is then "
"validated against those Checks.\n\n"
)
for item in summary.get("not_applicable", []):
text += (
f"Recommendation {item['recommendation']} is treated as not applicable: "
f"{item['reason']}\n"
)
return text + ("\n" if summary.get("not_applicable") else "")
def to_generate_response(
interaction: Interaction, roles: Mapping[str, RoleSpec]
) -> dict[str, Any]:
@@ -86,7 +105,7 @@ def to_generate_response(
"requirements_doc": requirements,
"architecture_doc": content(DeliverableKind.ARCHITECTURE_DESCRIPTION),
"mermaid_diagram": attachment(
DeliverableKind.ARCHITECTURE_DIAGRAM, "architecture.mmd"
DeliverableKind.TARGET_ARCHITECTURE, "architecture.mmd"
),
"terraform_code": attachment(
DeliverableKind.DEPLOYMENT_GUIDANCE, "terraform/main.tf"

View File

@@ -8,6 +8,7 @@ Execution store (review candidate 03) replaces them.
from __future__ import annotations
import asyncio
import json
import uuid
from collections import defaultdict
from collections.abc import Mapping, Sequence
@@ -42,12 +43,22 @@ class ExecutionRecorder(Protocol):
details: Mapping[str, Any],
) -> None: ...
async def record_lesson_candidates(
self, execution_id: str, record: Mapping[str, Any]
) -> None: ...
class InMemoryRecorder:
def __init__(self) -> None:
self.events: dict[str, list[LedgerEvent]] = defaultdict(list)
self.submissions: dict[str, list[dict[str, Any]]] = defaultdict(list)
self.statuses: dict[str, list[dict[str, Any]]] = defaultdict(list)
self.lessons: dict[str, dict[str, Any]] = {}
async def record_lesson_candidates(
self, execution_id: str, record: Mapping[str, Any]
) -> None:
self.lessons[execution_id] = dict(record)
async def record_events(
self, execution_id: str, events: Sequence[LedgerEvent]
@@ -154,3 +165,15 @@ class DatabaseRecorder:
agent_name=SESSION_AGENT_NAME,
state_data={"latest_execution_id": execution_id, "status": status},
)
async def record_lesson_candidates(
self, execution_id: str, record: Mapping[str, Any]
) -> None:
await asyncio.to_thread(
self._db.save_artifact,
artifact_id=str(uuid.uuid4()),
execution_id=execution_id,
artifact_type="lesson_candidates",
file_path="lesson-candidates.json",
content=json.dumps(dict(record), ensure_ascii=False),
)

View File

@@ -6,6 +6,7 @@ Agent Skills specification (https://agentskills.io/specification).
from __future__ import annotations
import hashlib
import re
from collections.abc import Iterable
from dataclasses import dataclass
@@ -123,6 +124,15 @@ def read_skill(skill_dir: Path) -> SkillMeta:
return SkillMeta(name=name, description=description, path=skill_dir)
def skill_version(skill_dir: Path) -> str:
"""A short content hash of every file in the Skill, to say which Skill was loaded."""
digest = hashlib.sha256()
for path in sorted(p for p in skill_dir.rglob("*") if p.is_file()):
digest.update(path.relative_to(skill_dir).as_posix().encode())
digest.update(path.read_bytes())
return digest.hexdigest()[:12]
def discover_skills(directories: Iterable[Path]) -> dict[str, SkillMeta]:
"""Index every Skill that is an immediate child of one of `directories`.

View File

@@ -3,12 +3,58 @@
from __future__ import annotations
import operator
import re
from pathlib import Path
from typing import Annotated, Any, TypedDict
# A Customer case id becomes a directory name, so it is a short lowercase slug.
CUSTOMER_CASE = re.compile(r"^[a-z0-9](?:[a-z0-9-]{0,62}[a-z0-9])?$")
# Executions started without a Customer case are grouped here, so every Execution's
# files sit under a case.
UNASSIGNED_CASE = "unassigned"
def check_customer_case(customer_case: object) -> str:
"""The Customer case id, defaulting to UNASSIGNED_CASE; a ValueError when it is
not a safe slug."""
if customer_case is None:
return UNASSIGNED_CASE
if not isinstance(customer_case, str) or not CUSTOMER_CASE.match(customer_case):
raise ValueError(
"customer_case must be 1 to 64 lowercase letters, digits or hyphens, "
f"starting and ending with a letter or digit; got {customer_case!r}"
)
return customer_case
def case_dir(output_root: Path, customer_case: str) -> Path:
"""Everything generated for one Customer case."""
return output_root / "deliverables" / "cases" / customer_case
def execution_dir(output_root: Path, execution_id: str, customer_case: str) -> Path:
"""Where one Execution's files go, under its Customer case."""
return case_dir(output_root, customer_case) / "executions" / execution_id
def lessons_file(output_root: Path, execution_id: str, customer_case: str) -> Path:
return (
output_root / "lessons" / "candidates" / customer_case / f"{execution_id}.json"
)
def merge_by_key(
current: dict[str, dict[str, Any]] | None, update: dict[str, dict[str, Any]] | None
) -> dict[str, dict[str, Any]]:
"""Parallel Specialist tasks each write their own Deliverable's entry."""
return {**(current or {}), **(update or {})}
class ExecutionState(TypedDict, total=False):
execution_id: str
session_id: str
customer_case: str # the Customer case this Execution belongs to
intent_version: int # which statement of the case's intent started this Execution
workflow_request: str
pre_approved: bool
allow_cloud_read: bool | None
@@ -21,6 +67,11 @@ class ExecutionState(TypedDict, total=False):
pending_questions: list[dict[str, Any]] # {kind, questions} not yet asked
awaiting_questions: list[dict[str, Any]] # {kind, questions} asked, not answered
answers: dict[str, list[dict[str, Any]]] # kind -> [{questions, reply}]
drafts: dict[
str, dict[str, Any]
] # kind -> {content, attachments} awaiting the user
# kind -> working memory its tools keep across attempts (e.g. pillar advice)
working: Annotated[dict[str, dict[str, Any]], merge_by_key]
pending_permission: dict[str, Any] | None # {tier, requested_by}
prompt: dict[str, Any] | None # what the user is being asked
@@ -28,6 +79,7 @@ class ExecutionState(TypedDict, total=False):
# applied to the Ledger by the next Orchestrator turn.
outcomes: Annotated[list[dict[str, Any]], operator.add]
outcomes_applied: int
refusals: list[dict[str, Any]] # the Orchestrator's refused tool calls
events_recorded: int
turns: int

125
app/states/lessons.py Normal file
View File

@@ -0,0 +1,125 @@
"""Lesson candidates: what went wrong in an Execution and what was asked to fix it.
Collected once, when the Execution ends, from what the Execution already holds: the
Ledger's events, the Specialists' outcome records and the tool refusals. They are raw
material for a reviewed change to a Skill, never memory carried into another Execution.
"""
from __future__ import annotations
from collections.abc import Mapping, Sequence
from typing import Any
from app.states.ledger import Ledger
from app.workflows.chain import VERDICTS_FILE, validation_outcome
from app.workflows.deliverables import (
DEPENDENCY_GRAPH,
DeliverableKind,
PhaseId,
)
MESSAGE_LIMIT = 500
_UNFINISHED = {"budget_exhausted", "stopped"}
def refusal(
role: str, deliverable: str | None, tool: str, tier: str, message: str
) -> dict[str, Any]:
"""One tool refusal, as stored in Execution state.
A cloud-read tool's message may describe the user's environment, so only its
leading words are kept.
"""
if tier == "cloud_read":
message = message.split(":", 1)[0]
return {
"role": role,
"deliverable": deliverable,
"tool": tool,
"message": message[:MESSAGE_LIMIT],
}
def collect_lesson_candidates(
*,
execution_id: str,
customer_case: str,
status: str,
ledger: Ledger,
outcomes: Sequence[Mapping[str, Any]],
orchestrator_refusals: Sequence[Mapping[str, Any]],
role_skills: Mapping[str, Sequence[str]],
skill_versions: Mapping[str, str],
) -> dict[str, Any]:
candidates: list[dict[str, Any]] = []
for event in ledger.events:
if event.type == "reopened":
candidates.append(
{
"type": "revision",
"deliverable": event.subject,
"reason": event.detail,
"requested_by": event.actor,
}
)
elif event.type == "escalated":
candidates.append(
{
"type": "revision_limit",
"deliverable": event.subject,
"reason": event.detail,
}
)
elif event.type == "changes_requested":
candidates.append(
{
"type": "change_request",
"phase": event.subject,
"deliverables": [
kind.value
for kind, spec in DEPENDENCY_GRAPH.items()
if spec.phase is PhaseId(event.subject)
],
"feedback": event.detail,
}
)
elif event.type == "released" and event.detail in _UNFINISHED:
candidates.append(
{
"type": "unfinished",
"deliverable": event.subject,
"role": event.actor,
"reason": event.detail,
}
)
for record in outcomes:
for item in record.get("refusals", ()):
candidates.append({"type": "refusal", **item})
if (
record.get("type") == "submitted"
and record.get("kind") == DeliverableKind.VALIDATION_RESULTS.value
):
outcome = validation_outcome(
record.get("attachments", {}).get(VERDICTS_FILE, "")
)
attempt = record.get("attempt")
for kind, items in (
("failed_check", outcome["failed"]),
("unverifiable_check", outcome["not_verifiable"]),
("proposed_addition", outcome["proposed_additions"]),
):
for item in items:
candidates.append({"type": kind, "attempt": attempt, **item})
for item in orchestrator_refusals:
candidates.append({"type": "refusal", **item})
return {
"execution_id": execution_id,
"customer_case": customer_case,
"status": status,
"role_skills": {role: list(skills) for role, skills in role_skills.items()},
"skill_versions": dict(skill_versions),
"candidates": candidates,
}

View File

@@ -28,8 +28,17 @@ from app.workflows.deliverables import (
check_submission,
deliverables_in,
)
from app.workflows.target_architecture import (
DIAGRAM_ATTACHMENT,
GRAPH_ATTACHMENT,
GraphError,
TargetArchitecture,
parse_graph,
render_mermaid,
)
ORCHESTRATOR = "orchestrator"
_TARGET = DeliverableKind.TARGET_ARCHITECTURE
CLOUD_READ = "cloud_read"
_NO_ARGS: Mapping[str, Any] = {"type": "object", "properties": {}}
@@ -186,10 +195,23 @@ class OrchestratorToolbox:
"summary": summary,
"deliverables": [kind.value for kind in deliverables_in(phase_id)],
}
graph = self._target_architecture()
if graph and phase_id is DEPENDENCY_GRAPH[_TARGET].phase:
# What the user is approving before any code is written.
self.approval["target_architecture"] = graph.summary()
return ToolResult(
f"Asked the user to approve {phase_id}.", TurnEnded("request_approval")
)
def _target_architecture(self) -> TargetArchitecture | None:
deliverable = self._deliverables.get(_TARGET.value)
if not deliverable:
return None
try:
return parse_graph(deliverable.get("attachments", {})[GRAPH_ATTACHMENT])
except (KeyError, GraphError):
return None
async def reopen(self, kind: str, reason: str) -> ToolResult:
try:
result = self.ledger.reopen(
@@ -231,8 +253,13 @@ def specialist_tools(
kind: DeliverableKind,
extra: Sequence[ToolSpec],
allow_cloud_read: bool | None,
upstream: Mapping[str, Mapping[str, Any]] | None = None,
) -> list[ToolSpec]:
"""A Specialist's tools: submit, ask, and its role's tools with the cloud-read gate."""
"""A Specialist's tools: submit, ask, and its role's tools with the cloud-read gate.
`upstream` is the Deliverables this one builds on; with it, a submission's links
along the Requirements chain are checked.
"""
async def submit_deliverable(content: str, attachments: Any = None) -> ToolResult:
try:
@@ -245,18 +272,34 @@ def specialist_tools(
"attachment names must be relative paths without '..': "
+ ", ".join(unsafe)
)
missing = check_submission(kind, content, attachments)
missing = check_submission(kind, content, attachments, upstream)
if missing:
return ToolResult(
"Not submitted. The Deliverable is missing: " + "; ".join(missing)
)
if kind is DeliverableKind.TARGET_ARCHITECTURE:
# The diagram is a view of the graph, so the engine draws it.
attachments[DIAGRAM_ATTACHMENT] = render_mermaid(
parse_graph(attachments[GRAPH_ATTACHMENT])
)
return ToolResult(f"Submitted {kind}.", Submitted(content, attachments))
async def needs_input(questions: Sequence[str]) -> ToolResult:
async def needs_input(
questions: Sequence[str], draft: str | None = None, attachments: Any = None
) -> ToolResult:
questions = tuple(q for q in questions if q)
if not questions:
return _refused("ask at least one question")
return ToolResult("Questions sent to the user.", NeedsInput(questions))
try:
draft_attachments = files_from_argument(attachments)
except TypeError as exc:
return _refused(f"attachments: {exc}")
return ToolResult(
"Questions sent to the user.",
NeedsInput(
questions, draft=draft or None, draft_attachments=draft_attachments
),
)
return [
ToolSpec(
@@ -274,9 +317,15 @@ def specialist_tools(
),
ToolSpec(
"needs_input",
"Ask the user questions you can't proceed without. Ends your turn.",
"Ask the user questions you can't proceed without. Ends your turn. When "
"you are asking them to confirm a draft, pass it as `draft` (with its "
"`attachments`): it is returned to you with their answers.",
_schema(
{"questions": {"type": "array", "items": {"type": "string"}}},
{
"questions": {"type": "array", "items": {"type": "string"}},
"draft": {"type": "string"},
"attachments": FILES_SCHEMA,
},
["questions"],
),
needs_input,

View File

@@ -11,7 +11,7 @@ import os
import re
import tempfile
from collections.abc import Awaitable, Callable, Iterable, Mapping
from dataclasses import dataclass
from dataclasses import dataclass, field
from pathlib import Path, PurePosixPath
from typing import Any
@@ -76,7 +76,11 @@ async def mermaid_check(diagram: str) -> ToolResult:
class TerraformRunner:
binary: str = "terraform"
timeout_seconds: float = 300
# Providers are downloaded here once and reused by every later check. Without it,
# each validate or plan downloads its providers again into a scratch directory.
plugin_cache_dir: str | None = None
# Terraform's plugin cache is not safe for concurrent installs.
_init_lock: asyncio.Lock = field(default_factory=asyncio.Lock, repr=False)
async def fmt(self, files: Mapping[str, str]) -> ToolResult:
refusal = _check_files(files)
@@ -147,9 +151,12 @@ class TerraformRunner:
)
async def _init(self, root: Path) -> ToolResult | None:
code, output = await self._run(
[self.binary, "init", "-backend=false", "-input=false", "-no-color"], root
)
command = [self.binary, "init", "-backend=false", "-input=false", "-no-color"]
if self.plugin_cache_dir:
async with self._init_lock:
code, output = await self._run(command, root)
else:
code, output = await self._run(command, root)
if code != 0:
return ToolResult(_clip(f"terraform init failed:\n{output}"))
return None
@@ -159,7 +166,12 @@ class TerraformRunner:
) -> tuple[int, str]:
environment = {**os.environ, "TF_IN_AUTOMATION": "1", **(env or {})}
if self.plugin_cache_dir:
environment["TF_PLUGIN_CACHE_DIR"] = self.plugin_cache_dir
cache = Path(self.plugin_cache_dir).expanduser().resolve()
cache.mkdir(parents=True, exist_ok=True)
environment["TF_PLUGIN_CACHE_DIR"] = str(cache)
# Each check runs in a scratch directory with no dependency lock file.
# Without this, Terraform downloads a cached provider again to verify it.
environment["TF_PLUGIN_CACHE_MAY_BREAK_DEPENDENCY_LOCK_FILE"] = "true"
try:
process = await asyncio.create_subprocess_exec(
*args,

View File

@@ -1,25 +1,31 @@
"""The design advisor's fan-out to one advisor per architecture-framework pillar.
Pillar advice is working material for the design recommendations Deliverable, not a
Deliverable itself, so it never touches the Ledger.
Deliverable itself, so it never touches the Ledger. It is kept in that Deliverable's
working memory, so a Revision re-consults only the pillars its feedback touches.
"""
from __future__ import annotations
import asyncio
import dataclasses
from collections.abc import Mapping, Sequence
from collections.abc import Mapping, MutableMapping, Sequence
from contextlib import AsyncExitStack
from typing import Any
from app.adk.runtime import (
AgentRuntime,
AgentTask,
RoleSpec,
RunLimiter,
Submitted,
ToolResult,
ToolSpec,
)
CONSULT_TOOL = "consult_pillar_advisors"
NO_GUIDANCE = "No guidance"
ADVICE_KEY = "pillar_advice" # working memory: pillar -> the advice last given
def pillar_consultation_tool(
@@ -27,8 +33,22 @@ def pillar_consultation_tool(
advisor: RoleSpec,
pillar_skills: Mapping[str, str],
extra_tools: Sequence[ToolSpec] = (),
memory: MutableMapping[str, Any] | None = None,
limiter: RunLimiter | None = None,
) -> ToolSpec:
"""A tool that runs one pillar advisor per pillar, in parallel, each with its Skill."""
"""A tool that runs one pillar advisor per pillar, in parallel, each with its Skill.
With `memory`, advice is remembered: naming `pillars` re-consults only those and
returns the remembered advice for the rest.
"""
async def advise(pillar: str, brief: str) -> str:
async with AsyncExitStack() as stack:
if limiter:
await stack.enter_async_context(limiter.slot())
return await _advise(
runtime, advisor, pillar, pillar_skills[pillar], brief, extra_tools
)
async def consult(brief: str, pillars: Sequence[str] | None = None) -> ToolResult:
chosen = list(pillars) if pillars else list(pillar_skills)
@@ -38,20 +58,41 @@ def pillar_consultation_tool(
f"Refused: unknown pillars {', '.join(unknown)}; "
f"choose from {', '.join(pillar_skills)}"
)
results = await asyncio.gather(
*(
_advise(
runtime, advisor, pillar, pillar_skills[pillar], brief, extra_tools
async with AsyncExitStack() as stack:
if limiter:
# The calling Specialist only waits here, so its slot goes to advisors.
await stack.enter_async_context(limiter.lend())
fresh = dict(
zip(
chosen,
await asyncio.gather(*(advise(pillar, brief) for pillar in chosen)),
)
for pillar in chosen
)
remembered: dict[str, str] = (
dict(memory.get(ADVICE_KEY, {})) if memory is not None else {}
)
return ToolResult("\n\n".join(results))
sections = []
for pillar in pillar_skills:
if pillar in fresh:
advice = fresh[pillar]
if not advice.startswith(NO_GUIDANCE):
remembered[pillar] = advice
sections.append(f"## {pillar}\n\n{advice}")
elif pillar in remembered:
sections.append(
f"## {pillar} (from your earlier consultation)\n\n"
f"{remembered[pillar]}"
)
if memory is not None:
memory[ADVICE_KEY] = remembered
return ToolResult("\n\n".join(sections))
return ToolSpec(
CONSULT_TOOL,
"Get guidance from one advisor per architecture-framework pillar, in parallel. "
"Give a brief with the workload, its requirements and the selected products.",
"Give a brief with the workload, its requirements and the selected products. "
"Omit `pillars` to consult every pillar afresh. Name `pillars` to re-consult "
"only those; advice already given for the others is returned unchanged.",
{
"type": "object",
"properties": {
@@ -84,12 +125,10 @@ async def _advise(
try:
outcome = await runtime.run(role, task, [_submit_advice_tool(), *extra_tools])
except Exception as exc: # noqa: BLE001 - one failed advisor must not lose the others
return f"## {pillar}\n\nNo guidance: the advisor failed ({exc})."
return f"{NO_GUIDANCE}: the advisor failed ({exc})."
if isinstance(outcome, Submitted):
return f"## {pillar}\n\n{outcome.content.strip()}"
return (
f"## {pillar}\n\nNo guidance: the advisor ended with {type(outcome).__name__}."
)
return outcome.content.strip()
return f"{NO_GUIDANCE}: the advisor ended with {type(outcome).__name__}."
def _submit_advice_tool() -> ToolSpec:

View File

@@ -7,10 +7,11 @@ Cloud-read gating is applied later, per Execution, by `specialist_tools`.
from __future__ import annotations
from collections.abc import Mapping, Sequence
from collections.abc import Mapping, MutableMapping, Sequence
from typing import Any
from app.adk.roles import PILLAR_ADVISOR
from app.adk.runtime import AgentRuntime, RoleSpec, ToolFactory, ToolSpec
from app.adk.runtime import AgentRuntime, RoleSpec, RunLimiter, ToolFactory, ToolSpec
from app.tools.mcp import McpCatalog
from app.tools.pillars import pillar_consultation_tool
from app.workflows.deliverables import DeliverableKind
@@ -23,26 +24,41 @@ def make_tool_factory(
*,
local_tools: Sequence[ToolSpec] = (),
catalog: McpCatalog | None = None,
limiter: RunLimiter | None = None,
) -> ToolFactory:
local_by_name = {tool.name: tool for tool in local_tools}
def mcp_tools(role: RoleSpec) -> list[ToolSpec]:
return catalog.tools_for(role.mcp_servers) if catalog else []
consult = None
advisor = roles.get(PILLAR_ADVISOR)
if pillar_skills and advisor:
# Pillar advisors can't ask for permission, so they get general-tier tools only.
advisor_tools = [tool for tool in mcp_tools(advisor) if tool.tier == "general"]
consult = pillar_consultation_tool(
runtime, advisor, pillar_skills, extra_tools=advisor_tools
)
# Pillar advisors can't ask for permission, so they get general-tier tools only.
advisor_tools = (
[tool for tool in mcp_tools(advisor) if tool.tier == "general"]
if advisor
else []
)
def tools_for(role: RoleSpec, deliverable: str | None) -> Sequence[ToolSpec]:
def tools_for(
role: RoleSpec, deliverable: str | None, working: MutableMapping[str, Any]
) -> Sequence[ToolSpec]:
tools = [local_by_name[name] for name in role.tools if name in local_by_name]
tools += mcp_tools(role)
if consult and deliverable == DeliverableKind.DESIGN_RECOMMENDATIONS.value:
tools.append(consult)
if (
pillar_skills
and advisor
and deliverable == DeliverableKind.DESIGN_RECOMMENDATIONS.value
):
tools.append(
pillar_consultation_tool(
runtime,
advisor,
pillar_skills,
extra_tools=advisor_tools,
memory=working,
limiter=limiter,
)
)
return tools
return tools_for

336
app/workflows/chain.py Normal file
View File

@@ -0,0 +1,336 @@
"""The data files along the Requirements chain, and what each must hold (ADR-0005).
requirements.json requirements with ids
components.json components, and the requirements each serves
products.json the product selected for each component
recommendations.json design recommendations with ids
target-architecture.json the graph (see target_architecture.py)
implementation.json which code satisfies each Check
verdicts.json the validation verdict on each Check
Each checker returns what is wrong, one entry per fault, so a Specialist can fix it
and submit again. `upstream` is the Deliverables a submission builds on, by kind; a
link to one of them is checked only when it is there.
"""
from __future__ import annotations
import json
from collections.abc import Mapping, Sequence
from typing import Any
from app.workflows.target_architecture import (
GRAPH_ATTACHMENT,
GraphError,
TargetArchitecture,
chain_gaps,
parse_graph,
)
REQUIREMENTS_FILE = "requirements.json"
COMPONENTS_FILE = "components.json"
PRODUCTS_FILE = "products.json"
RECOMMENDATIONS_FILE = "recommendations.json"
IMPLEMENTATION_FILE = "implementation.json"
VERDICTS_FILE = "verdicts.json"
REQUIREMENT_KINDS = ("functional", "non_functional")
VERDICTS = ("pass", "fail", "not_verifiable")
DRY_RUN_VERDICTS = ("pass", "fail", "not_run")
Upstream = Mapping[str, Mapping[str, Any]]
def read_list(upstream: Upstream | None, kind: str, name: str) -> list[dict[str, Any]]:
"""An upstream data file as a list of objects; empty when absent or unreadable."""
text = ((upstream or {}).get(kind) or {}).get("attachments", {}).get(name, "")
try:
data = json.loads(text)
except ValueError:
return []
return (
[item for item in data if isinstance(item, dict)]
if isinstance(data, list)
else []
)
def read_graph(upstream: Upstream | None) -> TargetArchitecture | None:
text = (
((upstream or {}).get("target_architecture") or {})
.get("attachments", {})
.get(GRAPH_ATTACHMENT, "")
)
try:
return parse_graph(text)
except GraphError:
return None
def _objects(
text: str, name: str, strings: Sequence[str], lists: Sequence[str] = ()
) -> tuple[list[dict[str, Any]], list[str]]:
"""The file's objects that have every required field, and what is wrong with it."""
try:
data = json.loads(text)
except ValueError as exc:
return [], [f"{name}: not valid JSON ({exc})"]
if not isinstance(data, list):
return [], [f"{name}: must be a JSON list"]
accepted, problems = [], []
for index, item in enumerate(data):
where = f"{name}[{index}]"
if not isinstance(item, dict):
problems.append(f"{where}: must be an object")
continue
missing = [
field
for field in strings
if not isinstance(item.get(field), str) or not item[field].strip()
]
missing += [
field
for field in lists
if not isinstance(item.get(field), list)
or not all(isinstance(value, str) for value in item[field])
]
if missing:
problems.append(f"{where}: needs {', '.join(missing)}")
continue
accepted.append(item)
return accepted, problems
def _repeated(
items: Sequence[Mapping[str, Any]], name: str, key: str = "id"
) -> list[str]:
seen: set[str] = set()
problems = []
for item in items:
if item[key] in seen:
problems.append(f"{name}: {key} {item[key]} is used more than once")
seen.add(item[key])
return problems
def check_requirements(
attachments: Mapping[str, str], upstream: Upstream | None
) -> list[str]:
name = REQUIREMENTS_FILE
items, problems = _objects(attachments[name], name, ("id", "kind", "text"))
problems += _repeated(items, name)
problems += [
f"{name}: {item['id']} kind must be one of {', '.join(REQUIREMENT_KINDS)}"
for item in items
if item["kind"] not in REQUIREMENT_KINDS
]
if not items and not problems:
problems.append(f"{name}: lists no requirements")
return problems
def check_components(
attachments: Mapping[str, str], upstream: Upstream | None
) -> list[str]:
name = COMPONENTS_FILE
items, problems = _objects(
attachments[name], name, ("id", "name"), ("requirements",)
)
problems += _repeated(items, name)
if not items and not problems:
problems.append(f"{name}: lists no components")
requirements = read_list(upstream, "requirements", REQUIREMENTS_FILE)
if requirements:
known = {item["id"] for item in requirements}
served: set[str] = set()
for item in items:
served.update(item["requirements"])
problems += [
f"{name}: {item['id']} serves {reference}, which is not a requirement"
for reference in item["requirements"]
if reference not in known
]
problems += [
f"{name}: functional requirement {item['id']} is served by no component"
for item in requirements
if item.get("kind") == "functional" and item["id"] not in served
]
return problems
def check_products(
attachments: Mapping[str, str], upstream: Upstream | None
) -> list[str]:
name = PRODUCTS_FILE
items, problems = _objects(attachments[name], name, ("component", "product"))
components = read_list(upstream, "technical_decomposition", COMPONENTS_FILE)
if components:
known = {item["id"] for item in components}
chosen = {item["component"] for item in items}
problems += [
f"{name}: {item['component']} is not a component"
for item in items
if item["component"] not in known
]
problems += [
f"{name}: component {identifier} has no product selected"
for identifier in sorted(known - chosen)
]
return problems
def check_recommendations(
attachments: Mapping[str, str], upstream: Upstream | None
) -> list[str]:
name = RECOMMENDATIONS_FILE
items, problems = _objects(
attachments[name], name, ("id", "pillar", "text"), ("components",)
)
problems += _repeated(items, name)
if not items and not problems:
problems.append(f"{name}: lists no recommendations")
components = read_list(upstream, "technical_decomposition", COMPONENTS_FILE)
if components:
known = {item["id"] for item in components}
problems += [
f"{name}: {item['id']} names {reference}, which is not a component"
for item in items
for reference in item["components"]
if reference not in known
]
return problems
def check_target_architecture(
attachments: Mapping[str, str], upstream: Upstream | None
) -> list[str]:
try:
graph = parse_graph(attachments[GRAPH_ATTACHMENT])
except GraphError as exc:
return list(exc.problems)
if upstream is None:
return []
return list(
chain_gaps(
graph,
requirements=read_list(upstream, "requirements", REQUIREMENTS_FILE),
components=read_list(upstream, "technical_decomposition", COMPONENTS_FILE),
products=read_list(upstream, "product_selection", PRODUCTS_FILE),
recommendations=read_list(
upstream, "design_recommendations", RECOMMENDATIONS_FILE
),
)
)
def _covers_every_check(
items: Sequence[Mapping[str, Any]], name: str, graph: TargetArchitecture | None
) -> list[str]:
problems = _repeated(items, name, "check")
if graph is None:
return problems
given = {item["check"] for item in items}
problems += [
f"{name}: {identifier} is not a Check in the Target architecture"
for identifier in sorted(given - set(graph.check_ids))
]
problems += [
f"{name}: Check {identifier} is not accounted for"
for identifier in graph.check_ids
if identifier not in given
]
return problems
def check_implementation(
attachments: Mapping[str, str], upstream: Upstream | None
) -> list[str]:
name = IMPLEMENTATION_FILE
items, problems = _objects(attachments[name], name, ("check", "file", "address"))
problems += _covers_every_check(items, name, read_graph(upstream))
problems += [
f"{name}: {item['check']} names {item['file']}, which is not attached"
for item in items
if item["file"] not in attachments
]
return problems
def check_verdicts(
attachments: Mapping[str, str], upstream: Upstream | None
) -> list[str]:
name = VERDICTS_FILE
try:
data = json.loads(attachments[name])
except ValueError as exc:
return [f"{name}: not valid JSON ({exc})"]
if not isinstance(data, dict):
return [
f"{name}: must be an object with dry_run, checks and proposed_additions"
]
problems: list[str] = []
dry_run = data.get("dry_run")
if (
not isinstance(dry_run, dict)
or dry_run.get("verdict") not in DRY_RUN_VERDICTS
or not isinstance(dry_run.get("evidence"), str)
or not dry_run["evidence"].strip()
):
problems.append(
f"{name}.dry_run: needs a verdict ({', '.join(DRY_RUN_VERDICTS)}) and evidence"
)
items, faults = _objects(
json.dumps(data.get("checks", [])),
f"{name}.checks",
("check", "verdict", "evidence"),
)
problems += faults
problems += [
f"{name}.checks: {item['check']} verdict must be one of {', '.join(VERDICTS)}"
for item in items
if item["verdict"] not in VERDICTS
]
problems += _covers_every_check(items, f"{name}.checks", read_graph(upstream))
_, faults = _objects(
json.dumps(data.get("proposed_additions", [])),
f"{name}.proposed_additions",
("statement", "applies_to", "reason"),
)
return problems + faults
def validation_outcome(verdicts_text: str) -> dict[str, Any]:
"""What a validation decided, for routing a Revision and for scoring.
Failed Checks are faults in the code. A Check that could not be verified and a
proposed addition are gaps in the Target architecture.
"""
try:
data = json.loads(verdicts_text)
except ValueError:
data = {}
if not isinstance(data, dict):
data = {}
checks = [item for item in data.get("checks", []) if isinstance(item, dict)]
def with_verdict(verdict: str) -> list[dict[str, Any]]:
return [
{"check": item.get("check"), "evidence": item.get("evidence")}
for item in checks
if item.get("verdict") == verdict
]
dry_run = data.get("dry_run") if isinstance(data.get("dry_run"), dict) else {}
return {
"dry_run": dry_run,
"passed": [
item.get("check") for item in checks if item.get("verdict") == "pass"
],
"failed": with_verdict("fail"),
"not_verifiable": with_verdict("not_verifiable"),
"proposed_additions": [
item
for item in data.get("proposed_additions", [])
if isinstance(item, dict)
],
}

View File

@@ -9,14 +9,34 @@ solution-guide template by suffix (e.g. "Products and features mapping" matches
from __future__ import annotations
import re
from collections.abc import Mapping
from collections.abc import Callable, Mapping
from dataclasses import dataclass
from enum import StrEnum
from app.workflows.chain import (
COMPONENTS_FILE,
IMPLEMENTATION_FILE,
PRODUCTS_FILE,
RECOMMENDATIONS_FILE,
REQUIREMENTS_FILE,
VERDICTS_FILE,
Upstream,
check_components,
check_implementation,
check_products,
check_recommendations,
check_requirements,
check_target_architecture,
check_verdicts,
)
from app.workflows.target_architecture import GRAPH_ATTACHMENT
class PhaseId(StrEnum):
REQUIREMENTS_DISCOVERY = "requirements_discovery"
# Design: ends with the Target architecture, approved before any code is written.
SOLUTION_ARCHITECTURE = "solution_architecture"
SOLUTION_IMPLEMENTATION = "solution_implementation"
SOLUTION_VALIDATION = "solution_validation"
PACKAGING = "packaging"
@@ -28,9 +48,9 @@ class DeliverableKind(StrEnum):
REQUIREMENTS = "requirements"
TECHNICAL_DECOMPOSITION = "technical_decomposition"
PRODUCT_SELECTION = "product_selection"
ARCHITECTURE_DIAGRAM = "architecture_diagram"
ARCHITECTURE_DESCRIPTION = "architecture_description"
DESIGN_RECOMMENDATIONS = "design_recommendations"
TARGET_ARCHITECTURE = "target_architecture"
ARCHITECTURE_DESCRIPTION = "architecture_description"
DEPLOYMENT_GUIDANCE = "deployment_guidance"
VALIDATION_RESULTS = "validation_results"
SOLUTION_GUIDE = "solution_guide"
@@ -45,6 +65,13 @@ class DeliverableSpec:
sections: tuple[str, ...] = ()
attachments: tuple[str, ...] = ()
def contract(self) -> dict[str, list[str]]:
"""What a submission must contain, as plain data for the Specialist's context."""
return {
"sections": list(self.sections),
"attachments": list(self.attachments),
}
_K = DeliverableKind
@@ -61,12 +88,15 @@ DEPENDENCY_GRAPH: Mapping[DeliverableKind, DeliverableSpec] = {
"Current state",
"Dependencies",
),
attachments=(REQUIREMENTS_FILE,),
),
DeliverableSpec(
_K.TECHNICAL_DECOMPOSITION,
PhaseId.REQUIREMENTS_DISCOVERY,
"decomposer",
depends_on=(_K.REQUIREMENTS,),
sections=("Components",),
attachments=(COMPONENTS_FILE,),
),
DeliverableSpec(
_K.PRODUCT_SELECTION,
@@ -74,19 +104,7 @@ DEPENDENCY_GRAPH: Mapping[DeliverableKind, DeliverableSpec] = {
"product-selector",
depends_on=(_K.TECHNICAL_DECOMPOSITION,),
sections=("Products and features mapping",),
),
DeliverableSpec(
_K.ARCHITECTURE_DIAGRAM,
PhaseId.SOLUTION_ARCHITECTURE,
"diagrammer",
depends_on=(_K.PRODUCT_SELECTION,),
attachments=("architecture.mmd",),
),
DeliverableSpec(
_K.ARCHITECTURE_DESCRIPTION,
PhaseId.SOLUTION_ARCHITECTURE,
"describer",
depends_on=(_K.ARCHITECTURE_DIAGRAM, _K.PRODUCT_SELECTION),
attachments=(PRODUCTS_FILE,),
),
DeliverableSpec(
_K.DESIGN_RECOMMENDATIONS,
@@ -101,17 +119,38 @@ DEPENDENCY_GRAPH: Mapping[DeliverableKind, DeliverableSpec] = {
"Performance efficiency",
"Sustainability",
),
attachments=(RECOMMENDATIONS_FILE,),
),
DeliverableSpec(
_K.TARGET_ARCHITECTURE,
PhaseId.SOLUTION_ARCHITECTURE,
"target-architect",
depends_on=(_K.PRODUCT_SELECTION, _K.DESIGN_RECOMMENDATIONS),
sections=("Elements", "Relationships", "Checks"),
# The engine adds architecture.mmd, rendered from the graph, on submission.
attachments=(GRAPH_ATTACHMENT,),
),
DeliverableSpec(
_K.ARCHITECTURE_DESCRIPTION,
PhaseId.SOLUTION_IMPLEMENTATION,
"describer",
depends_on=(_K.TARGET_ARCHITECTURE,),
sections=("Components", "Flow"),
),
DeliverableSpec(
_K.DEPLOYMENT_GUIDANCE,
PhaseId.SOLUTION_ARCHITECTURE,
PhaseId.SOLUTION_IMPLEMENTATION,
"iac-author",
depends_on=(_K.PRODUCT_SELECTION, _K.DESIGN_RECOMMENDATIONS),
depends_on=(_K.TARGET_ARCHITECTURE,),
sections=(
"Deployment prerequisites",
"Step-by-step deployment instructions",
),
attachments=("terraform/main.tf", "terraform/variables.tf"),
attachments=(
"terraform/main.tf",
"terraform/variables.tf",
IMPLEMENTATION_FILE,
),
),
DeliverableSpec(
_K.VALIDATION_RESULTS,
@@ -119,6 +158,7 @@ DEPENDENCY_GRAPH: Mapping[DeliverableKind, DeliverableSpec] = {
"validator",
depends_on=(_K.DEPLOYMENT_GUIDANCE,),
sections=("Deployment dry-run", "Architecture and policy analysis"),
attachments=(VERDICTS_FILE,),
),
DeliverableSpec(
_K.SOLUTION_GUIDE,
@@ -139,6 +179,20 @@ DEPENDENCY_GRAPH: Mapping[DeliverableKind, DeliverableSpec] = {
)
}
# What each data file along the Requirements chain must hold, checked at submission.
DATA_CHECKS: Mapping[
DeliverableKind,
tuple[str, Callable[[Mapping[str, str], Upstream | None], list[str]]],
] = {
_K.REQUIREMENTS: (REQUIREMENTS_FILE, check_requirements),
_K.TECHNICAL_DECOMPOSITION: (COMPONENTS_FILE, check_components),
_K.PRODUCT_SELECTION: (PRODUCTS_FILE, check_products),
_K.DESIGN_RECOMMENDATIONS: (RECOMMENDATIONS_FILE, check_recommendations),
_K.TARGET_ARCHITECTURE: (GRAPH_ATTACHMENT, check_target_architecture),
_K.DEPLOYMENT_GUIDANCE: (IMPLEMENTATION_FILE, check_implementation),
_K.VALIDATION_RESULTS: (VERDICTS_FILE, check_verdicts),
}
def inputs_of(kind: DeliverableKind) -> tuple[DeliverableKind, ...]:
"""The Deliverables a Specialist works from, in graph order.
@@ -179,9 +233,16 @@ def _normalise(text: str) -> str:
def check_submission(
kind: DeliverableKind, content: str, attachments: Mapping[str, str]
kind: DeliverableKind,
content: str,
attachments: Mapping[str, str],
upstream: Upstream | None = None,
) -> tuple[str, ...]:
"""Return what a submission lacks against its contract; empty means it's complete."""
"""Return what a submission lacks against its contract; empty means it's complete.
`upstream` is the Deliverables this one builds on, by kind. With it, the links
along the Requirements chain are checked too.
"""
if not content.strip():
return ("content: must not be empty",)
@@ -195,4 +256,10 @@ def check_submission(
for name in spec.attachments:
if name not in attachments:
missing.append(f"attachment: {name}")
elif not attachments[name].strip():
missing.append(f"attachment: {name} must not be empty")
if kind in DATA_CHECKS:
name, check = DATA_CHECKS[kind]
if attachments.get(name, "").strip():
missing.extend(check(attachments, upstream))
return tuple(missing)

View File

@@ -18,7 +18,7 @@ from typing import Any, cast
from langgraph.graph import END, START, StateGraph
from app.adk.runtime import AgentRuntime, RoleSpec, ToolFactory
from app.adk.runtime import AgentRuntime, RoleSpec, RunLimiter, ToolFactory
from app.nodes.execution_nodes import make_execution_nodes
from app.recorder import ExecutionRecorder
from app.states.execution_state import ExecutionState
@@ -32,6 +32,8 @@ def build_execution_graph(
output_root: Path,
max_orchestrator_turns: int = 100,
tool_factory: ToolFactory | None = None,
limiter: RunLimiter | None = None,
skill_versions: Mapping[str, str] | None = None,
) -> StateGraph:
nodes = make_execution_nodes(
runtime=runtime,
@@ -40,6 +42,8 @@ def build_execution_graph(
output_root=output_root,
max_orchestrator_turns=max_orchestrator_turns,
tool_factory=tool_factory,
limiter=limiter,
skill_versions=skill_versions,
)
graph = StateGraph(ExecutionState)

View File

@@ -0,0 +1,283 @@
"""The Target architecture: one explicit graph, established before code (ADR-0005).
Elements every component and the resource that realises it
Relationships how Elements connect
Checks the numbered, verifiable conditions each must meet, each traced
to the requirement or design recommendation it comes from
This module is pure: it parses the graph, states what is wrong with it, renders it as
a diagram, and says whether the Requirements chain through it is complete.
"""
from __future__ import annotations
import json
import re
from collections.abc import Iterable, Mapping, Sequence
from dataclasses import dataclass
from typing import Any
GRAPH_ATTACHMENT = "target-architecture.json"
DIAGRAM_ATTACHMENT = "architecture.mmd"
@dataclass(frozen=True)
class Element:
id: str
name: str
product: str
resource_type: str
component: str | None = None # None for a supporting Element, e.g. an identity
@dataclass(frozen=True)
class Relationship:
id: str
source: str # "from" in the data
target: str # "to" in the data
kind: str
@dataclass(frozen=True)
class Check:
id: str
applies_to: str # an Element or Relationship id
statement: str
source: str # a requirement or design recommendation id
@dataclass(frozen=True)
class NotApplicable:
recommendation: str
reason: str
@dataclass(frozen=True)
class TargetArchitecture:
elements: tuple[Element, ...]
relationships: tuple[Relationship, ...]
checks: tuple[Check, ...]
not_applicable: tuple[NotApplicable, ...] = ()
@property
def check_ids(self) -> tuple[str, ...]:
return tuple(check.id for check in self.checks)
def summary(self) -> dict[str, Any]:
"""What a user approving the design is told about the graph."""
return {
"elements": len(self.elements),
"relationships": len(self.relationships),
"checks": len(self.checks),
"not_applicable": [
{"recommendation": item.recommendation, "reason": item.reason}
for item in self.not_applicable
],
}
class GraphError(ValueError):
"""The graph is not well-formed; `problems` says how, one entry per fault."""
def __init__(self, problems: Sequence[str]) -> None:
super().__init__("; ".join(problems))
self.problems = tuple(problems)
def parse_graph(text: str) -> TargetArchitecture:
"""Parse target-architecture.json, or raise GraphError naming every fault."""
name = GRAPH_ATTACHMENT
try:
data = json.loads(text)
except ValueError as exc:
raise GraphError([f"{name}: not valid JSON ({exc})"]) from exc
if not isinstance(data, dict):
raise GraphError(
[f"{name}: must be an object with elements, relationships and checks"]
)
problems: list[str] = []
def items(key: str, required: Sequence[str]) -> list[dict[str, str]]:
raw = data.get(key, [])
if not isinstance(raw, list):
problems.append(f"{name}.{key}: must be a list")
return []
accepted = []
for index, item in enumerate(raw):
where = f"{name}.{key}[{index}]"
if not isinstance(item, dict):
problems.append(f"{where}: must be an object")
continue
missing = [
field
for field in required
if not isinstance(item.get(field), str) or not item[field].strip()
]
if missing:
problems.append(f"{where}: needs a non-empty {', '.join(missing)}")
continue
accepted.append(item)
return accepted
elements = tuple(
Element(
id=item["id"],
name=item["name"],
product=item["product"],
resource_type=item["resource_type"],
component=item.get("component") or None,
)
for item in items("elements", ("id", "name", "product", "resource_type"))
)
relationships = tuple(
Relationship(item["id"], item["from"], item["to"], item["kind"])
for item in items("relationships", ("id", "from", "to", "kind"))
)
checks = tuple(
Check(item["id"], item["applies_to"], item["statement"], item["source"])
for item in items("checks", ("id", "applies_to", "statement", "source"))
)
not_applicable = tuple(
NotApplicable(item["recommendation"], item["reason"])
for item in items("not_applicable", ("recommendation", "reason"))
)
if not elements and not problems:
problems.append(f"{name}.elements: the graph has no Elements")
seen: set[str] = set()
ids = [thing.id for group in (elements, relationships, checks) for thing in group]
for identifier in ids:
if identifier in seen:
problems.append(f"{name}: id {identifier} is used more than once")
seen.add(identifier)
element_ids = {element.id for element in elements}
for relationship in relationships:
for end, value in (("from", relationship.source), ("to", relationship.target)):
if value not in element_ids:
problems.append(
f"{name}: Relationship {relationship.id} {end} {value}, "
"which is not an Element"
)
targets = element_ids | {relationship.id for relationship in relationships}
for check in checks:
if check.applies_to not in targets:
problems.append(
f"{name}: Check {check.id} applies to {check.applies_to}, "
"which is not an Element or Relationship"
)
checked = {check.applies_to for check in checks}
for element in elements:
if element.id not in checked:
problems.append(f"{name}: Element {element.id} has no Check")
if problems:
raise GraphError(problems)
return TargetArchitecture(elements, relationships, checks, not_applicable)
# --- the Requirements chain through the graph ----------------------------------------
def chain_gaps(
graph: TargetArchitecture,
*,
requirements: Sequence[Mapping[str, Any]],
components: Sequence[Mapping[str, Any]],
products: Sequence[Mapping[str, Any]],
recommendations: Sequence[Mapping[str, Any]],
) -> tuple[str, ...]:
"""Every place the Requirements chain stops before reaching a Check.
A functional requirement is carried by a component that has an Element, or by a
Check of its own. A non-functional requirement and a design recommendation are
each a constraint, so each must be the source of a Check (a recommendation may
instead be marked not applicable, with a reason).
"""
gaps: list[str] = []
requirement_ids = {item["id"] for item in requirements}
recommendation_ids = {item["id"] for item in recommendations}
sources = {check.source for check in graph.checks}
realised = {element.component for element in graph.elements if element.component}
for check in graph.checks:
if check.source not in requirement_ids | recommendation_ids:
gaps.append(
f"Check {check.id} has source {check.source}, which is not a "
"requirement or design recommendation"
)
for element in graph.elements:
if element.component and element.component not in {c["id"] for c in components}:
gaps.append(
f"Element {element.id} realises {element.component}, "
"which is not a component"
)
for component in components:
if component["id"] not in realised:
gaps.append(f"Component {component['id']} has no Element")
for requirement in requirements:
if requirement["id"] in sources:
continue
served = any(
requirement["id"] in component.get("requirements", ())
and component["id"] in realised
for component in components
)
if requirement.get("kind") == "functional" and served:
continue
gaps.append(f"Requirement {requirement['id']} has no Check")
for selection in products:
used = any(
element.component == selection["component"]
and element.product == selection["product"]
for element in graph.elements
)
if not used:
gaps.append(
f"Product {selection['product']} selected for "
f"{selection['component']} appears on no Element"
)
excused = {item.recommendation for item in graph.not_applicable}
for item in graph.not_applicable:
if item.recommendation not in recommendation_ids:
gaps.append(
f"not_applicable names {item.recommendation}, "
"which is not a design recommendation"
)
for recommendation in recommendations:
if recommendation["id"] not in sources | excused:
gaps.append(f"Recommendation {recommendation['id']} has no Check")
return tuple(gaps)
# --- the diagram is a view of the graph ----------------------------------------------
_UNSAFE = re.compile(r'["\[\]{}()<>|`#;]')
def _label(*parts: str) -> str:
return "<br/>".join(_UNSAFE.sub(" ", part).strip() for part in parts if part)
def _node_ids(ids: Iterable[str]) -> dict[str, str]:
"""Mermaid-safe node ids; the graph's own ids may hold characters Mermaid rejects."""
return {identifier: f"n{index}" for index, identifier in enumerate(ids)}
def render_mermaid(graph: TargetArchitecture) -> str:
"""The architecture diagram: every Element and Relationship, exactly once."""
nodes = _node_ids(element.id for element in graph.elements)
lines = ["flowchart LR"]
for element in graph.elements:
lines.append(
f' {nodes[element.id]}["{_label(element.name, element.product)}"]'
)
for relationship in graph.relationships:
lines.append(
f" {nodes[relationship.source]} -->|{_label(relationship.kind)}| "
f"{nodes[relationship.target]}"
)
return "\n".join(lines) + "\n"