feat: all four evaluation cases pass; intake, checklist review, thresholds

Case 04 (interactive)
- A scripted answer may name who it is for, so facts written for
  requirements discovery are not given to a Specialist confirming a draft.
- A reply that restates what the agreed requirements already say is read
  as an answer, not a request for changes.
- With a user present, a first submission is refused until the user has
  been shown a draft of that Deliverable (CONFIRM_DRAFTS).

Routing and reviews
- Intake: a message that starts something is read first. A question is
  answered and starts nothing; a change to delivered work starts as a new
  request with a note, until Change requests are built.
- Requirements and recommendations list what infrastructure code can show
  (conditions) apart from what it cannot (outside_code); the latter needs
  no Check and validation proposes nothing for it.
- A requirements review, "names no products", behind its own switch (off).
- Each pillar advisor reviews its pillar's checklist; the engine writes the
  reviews into the design recommendations. The judged checklist check went
  from 9-45% a pillar to 100% in one run of case 04.

Evaluation
- Per-case thresholds for Revisions, seconds and tokens, set from every
  completed run on this build.
- A progress file after every agent run, so an interrupted run keeps its
  time and tokens.
- eval/calibrate_judge.py with a 32-question set awaiting a person's
  labels. Unlabelled: Jev stable at 1.00, Gemini 0.92, agreeing on 0.91.

Smaller
- An Execution that ends with warnings writes its Deliverables, with the
  warnings and each status in the manifest.
- CI installs the application's dependencies before linting and testing.
- A load-balancer pattern in the IaC patterns Skill.

Plans record nine runs of 2026-10-08. Iteration 6 is complete.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-10-09 09:32:57 +01:00
parent d41607f505
commit 654370bb46
53 changed files with 2122 additions and 106 deletions

View File

@@ -3,6 +3,10 @@
A new task starts an Execution; a message on an input-required task answers the
Execution's prompt. Approvals and permissions count only as structured data parts,
never as free text.
Before a new task starts anything its message is read (app/nodes/intake.py): a
question is answered and starts nothing, and a change to delivered work is started
as a new request with a note saying so, until Change requests are built.
"""
from __future__ import annotations
@@ -25,6 +29,7 @@ from google.protobuf import json_format
from app.agent import ExecutionEngine, Interaction
from app.nodes.execution_nodes import GUIDE_FILENAME
from app.nodes.intake import ABOUT, CHANGE_NOTE, CHANGE_REQUEST, QUESTION, Intake
from app.presentation import describe_prompt
from app.workflows.deliverables import DeliverableKind
@@ -34,14 +39,17 @@ STRUCTURED_REPLY_KEYS = ("approval", "permission")
class SolutionArchitectExecutor(AgentExecutor):
def __init__(self, engine: ExecutionEngine) -> None:
def __init__(self, engine: ExecutionEngine, intake: Intake | None = None) -> None:
self._engine = engine
self._intake = intake
async def execute(self, context: RequestContext, event_queue: EventQueue) -> None:
task_id = context.task_id or ""
updater = TaskUpdater(event_queue, task_id, context.context_id or "")
new = context.current_task is None
note: str | None = None
try:
if context.current_task is None:
if new:
history = [context.message] if context.message is not None else []
await event_queue.enqueue_event(
new_task(
@@ -53,6 +61,27 @@ class SolutionArchitectExecutor(AgentExecutor):
)
await updater.start_work()
options = _options(context)
# A pre-approved caller is a batch job: it sends work, not questions.
if self._intake and not options.get("pre_approved"):
reading = await self._intake.read(
context.get_user_input(),
options.get("customer_case"),
scope=task_id,
)
if reading.kind == QUESTION:
await updater.complete(
updater.new_agent_message(
[
new_text_part(ABOUT),
new_data_part(
{"status": "answered", "intake": QUESTION}
),
]
)
)
return
if reading.kind == CHANGE_REQUEST:
note = CHANGE_NOTE
interaction = await self._engine.start(
context.get_user_input(),
session_id=context.context_id,
@@ -76,7 +105,7 @@ class SolutionArchitectExecutor(AgentExecutor):
)
)
return
await self._publish(interaction, updater)
await self._publish(interaction, updater, note if new else None)
async def cancel(self, context: RequestContext, event_queue: EventQueue) -> None:
# The checkpoint stays; cancelling only ends the A2A task.
@@ -85,12 +114,15 @@ class SolutionArchitectExecutor(AgentExecutor):
)
await updater.cancel()
async def _publish(self, interaction: Interaction, updater: TaskUpdater) -> None:
async def _publish(
self, interaction: Interaction, updater: TaskUpdater, note: str | None = None
) -> None:
if interaction.status == "awaiting_user" and interaction.prompt is not None:
prompt = dict(interaction.prompt)
await updater.requires_input(
updater.new_agent_message(
[
*([new_text_part(note)] if note else []),
new_text_part(describe_prompt(prompt)),
new_data_part({"prompt": prompt}),
]