Files
nisse/nisse-dev/src/nisse_dev/views.py
T
dongarandClaude Opus 5.5 37965aefca
CI / checks (push) Skipped
CI / checks (pull_request) Successful in 2m17s
Fix the run-breaking tool-result order; e2e test runs the slice (M2.1, PR D)
The first live run of the new loop's e2e test found that a fact recorded
while start_milestone was in flight landed between the tool call and its
result; both API formats reject that, so every later request in the run
failed. Views now keep each result right after its call. Scripted test
models check the APIs' ordering rules, so this can't pass unit tests
again.

The e2e tests no longer depend on the sandbox's history: each run asks for
a uniquely named function and works against a base branch of its own,
deleted afterwards. The review step asks for a change that can't already
be there. The scripted operator gives up after five fruitless answers, and
each run keeps its state store for inspection. The driver's prompt says
review fixes stay in the current milestone.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE
2026-10-03 15:11:06 +00:00

159 lines
6.1 KiB
Python

"""Per-role views of a work item's transcript (architecture L6; Development
Plan M1.2).
The log records every role's events in one work item. A model must not read
another role's turns as its own: in the second live run the planner, shown
the executor's tool calls as assistant turns, took that work for its own
(#78). So what a role is shown is projected here, before rendering:
- its own events, unchanged;
- what the operator said, as a user message labelled with where it came from;
- the orchestrator's facts and gates, as labelled statements;
- another model role's work only where it needs it, and only as a labelled
report or narration, never as a turn.
The projection keeps each event's id, so cursors (``after``) and compaction
ranges still line up with the log. Everything not the viewer's own becomes a
``UserMessage``: from where the viewer sits, it is all input.
"""
from dataclasses import replace
from nisse_dev.canonical import (
AssistantText,
DelegationCompleted,
Fact,
GateOpened,
PlanSubmitted,
StoredEvent,
ToolCall,
ToolResult,
UserMessage,
)
# How much of another role's tool result a narration line keeps.
NARRATION_RESULT_CHARS = 300
# The roles whose work the viewer is shown, and how: "report" keeps only the
# handback, "narrate" keeps a line per step. Anything not listed is not shown.
_SHOWN: dict[str, dict[str, str]] = {
"planner": {},
"driver": {"helper": "report"},
"executor": {"helper": "report"},
"helper": {"executor": "narrate", "driver": "narrate"},
}
def operator_label(source: str) -> str:
"""Where an operator message came from, for its label."""
origin, _, idle = source.partition(" (sent while idle")
if origin.startswith("transport:"):
channel = "Matrix"
elif origin.startswith("gitea:pr:"):
number = origin.split(":")[2]
channel = f"comment on PR #{number}"
elif origin.startswith("gitea:issue-body:"):
channel = f"issue #{origin.split(':')[2]}"
elif origin.startswith("gitea:issue:"):
number = origin.split(":")[2]
channel = f"comment on plan issue #{number}"
elif origin.startswith("gitea:"):
channel = "Gitea comment"
else:
channel = origin or "unknown"
if idle:
channel += ", sent while idle"
return f"operator, {channel}"
def _said(stored: StoredEvent, label: str, text: str) -> StoredEvent:
return replace(
stored,
event=UserMessage(text=f"[{label}] {text}", source=f"view:{stored.actor}"),
)
def _narration(call: ToolCall, result: ToolResult | None) -> str:
target = call.arguments.get("path") or call.arguments.get("command")
step = f"{call.tool}({str(target)[:80]})" if target else call.tool
if result is None:
return f"{step} → no result recorded"
outcome = "ok" if result.success else "failed"
text = result.result.strip()
if len(text) > NARRATION_RESULT_CHARS:
text = text[:NARRATION_RESULT_CHARS] + " …"
return f"{step} → {outcome}: {text}" if text else f"{step} → {outcome}"
def project(events: list[StoredEvent], viewer: str) -> list[StoredEvent]:
"""The events as ``viewer`` should be shown them, in log order."""
shown = _SHOWN.get(viewer)
if shown is None:
# A role with no view defined (the expert is stateless, the
# classifier sees only an action) reads the log as it is, with
# results kept after their calls.
return _results_follow_calls(list(events))
results = {
e.event.call_id: e.event for e in events if isinstance(e.event, ToolResult)
}
out: list[StoredEvent] = []
for stored in events:
event, actor = stored.event, stored.actor
if isinstance(event, PlanSubmitted):
# A record for the executor's scope and the dump, not a turn:
# the planner's own submit_plan call already carries the plan.
if viewer == "helper":
out.append(_said(stored, "approved plan", event.plan))
continue
if actor == viewer:
out.append(stored)
elif actor == "operator" and isinstance(event, UserMessage):
out.append(_said(stored, operator_label(event.source), event.text))
elif isinstance(event, Fact):
out.append(_said(stored, "orchestrator", event.text))
elif isinstance(event, GateOpened):
out.append(
_said(stored, "orchestrator", f"gate {event.gate}: {event.reason}")
)
elif shown.get(actor) == "report" and isinstance(event, DelegationCompleted):
out.append(
_said(
stored,
f"{actor} report, {event.exit_status}",
event.handback,
)
)
elif shown.get(actor) == "narrate":
if isinstance(event, ToolCall):
line = _narration(event, results.get(event.call_id))
out.append(_said(stored, actor, line))
elif isinstance(event, AssistantText) and event.text.strip():
out.append(_said(stored, actor, event.text.strip()))
return _results_follow_calls(out)
def _results_follow_calls(events: list[StoredEvent]) -> list[StoredEvent]:
"""Keep each tool result right after its call, as both API formats
require: anything recorded while a call was in flight (a fact the
harness wrote while handling it, say) moves to just after the results.
A call with no result yet keeps its place; the renderer answers it."""
out: list[StoredEvent] = []
pending: set[str] = set()
held: list[StoredEvent] = []
for stored in events:
event = stored.event
if isinstance(event, ToolCall):
pending.add(event.call_id)
out.append(stored)
elif isinstance(event, ToolResult) and event.call_id in pending:
pending.discard(event.call_id)
out.append(stored)
if not pending:
out.extend(held)
held = []
elif pending:
held.append(stored)
else:
out.append(stored)
return out + held