feat(harness): taskdoc per-step generator with byte budget (D16)

Genera un documento compatto per fase, derivato da artefatti + effective_decisions,
con byte budget enforced (target <20k token per un 35B/<200k). Mai incorpora
physical.yaml (~190k token, fatale). D15+D16 complementari: il brief delle decisioni
e' effective-aware, quindi post-rollback riflette lo stato corretto (le stale di
fasi > current_phase sono escluse).
6 tests (question+schema, no physical.yaml, budget ok/violato, header fase,
stale-excluded post-rollback). 32 total passing.
This commit is contained in:
2026-06-26 22:38:24 +02:00
parent 4b584934c8
commit a579cb88e5
2 changed files with 139 additions and 0 deletions
+69
View File
@@ -0,0 +1,69 @@
"""Per-step task document generator (spec D16, §4.9).
Emette un singolo documento compatto per fase/step, derivato dagli artefatti precedenti
e dalla vista effective delle decisioni, con byte budget enforced (target <20k token
per un modello 35B/<200k). MAI incorpora physical.yaml (~190k token, fatale).
D15+D16 complementari: il task doc e' generato dalla vista effective_decisions, quindi
post-rollback riflette automaticamente lo stato corretto (le decisioni stale di fasi
> current_phase sono escluse).
"""
from __future__ import annotations
from dataclasses import dataclass
from pathlib import Path
from nsp.phase import effective_decisions
from nsp.workflow import load_workflow
MAX_BODY_BYTES = 80_000 # ~20k token (target per task document di una fase)
@dataclass
class TaskDoc:
phase: int
body: str
byte_budget_ok: bool
def generate_task_doc(
session_dir: Path | str,
phase: int,
promoted_tables: list[str] | None = None,
) -> TaskDoc:
"""Genera il documento di task per la fase `phase`.
Contenuto (compatti, mai artefatti integrali fatali):
- Domanda (question.md) se presente.
- Schema linking (schema_linking.json) solo da fase >= 4.
- Brief delle decisioni effective (esclude stale post-rollback, esclude ritirate).
- Header del task con il numero/nome della fase.
"""
session_dir = Path(session_dir)
parts: list[str] = []
q = session_dir / "question.md"
if q.exists():
parts.append("## Domanda\n" + q.read_text())
sl = session_dir / "schema_linking.json"
if sl.exists() and phase >= 4:
parts.append("## Schema linking (deciso)\n```json\n" + sl.read_text() + "\n```")
# Brief decisioni effective (D15-aware)
eff = effective_decisions(session_dir)
if eff:
lines = [f"- {d.type} | {d.subject} | {d.detail}" for d in eff]
parts.append("## Decisioni effettive (effective)\n" + "\n".join(lines))
# Header fase
try:
wf = load_workflow()
name = wf.phase_name(phase)
header = f"## Task: fase {phase} ({name})"
except Exception:
header = f"## Task: fase {phase}"
parts.append(header)
body = "\n\n".join(parts)
return TaskDoc(phase=phase, body=body, byte_budget_ok=len(body.encode()) <= MAX_BODY_BYTES)
+70
View File
@@ -0,0 +1,70 @@
from pathlib import Path
from nsp.taskdoc import generate_task_doc
def test_task_doc_includes_question_and_schema_scope(tmp_path):
s = tmp_path / "sess"; s.mkdir()
(s / "question.md").write_text("# Domanda\nQuanti pazienti?\n## Assunzioni\n- a")
(s / "schema_linking.json").write_text(
'{"question":"q","candidates":[{"kind":"table","name":"pazienti"}],"joins":[],"excluded":[],"open_questions":[]}'
)
doc = generate_task_doc(session_dir=s, phase=7)
assert "Quanti pazienti?" in doc.body
assert "pazienti" in doc.body
assert doc.byte_budget_ok is True
def test_task_doc_never_embeds_full_physical_yaml(tmp_path):
"""physical.yaml e' fatale per un 35B/<200k (~190k token). Mai incorporarlo."""
s = tmp_path / "sess"; s.mkdir()
(s / "question.md").write_text("q")
# un physical.yaml enorme fuori dalla sessione (come in ChironeWp3: artifacts/mschema/)
(s.parent / "physical.yaml").write_text("x: " + "y" * 800_000)
doc = generate_task_doc(session_dir=s, phase=7)
assert "physical.yaml" not in doc.body
assert len(doc.body) < 100_000 # bounded
def test_task_doc_byte_budget_enforced_on_normal_input(tmp_path):
s = tmp_path / "sess"; s.mkdir()
(s / "question.md").write_text("q")
doc = generate_task_doc(session_dir=s, phase=1)
assert doc.byte_budget_ok is True
def test_task_doc_byte_budget_violation_flagged(tmp_path):
"""Se un artefatto di sessione e' enorme (input perverso), byte_budget_ok diventa False."""
s = tmp_path / "sess"; s.mkdir()
(s / "question.md").write_text("q")
(s / "schema_linking.json").write_text("x: " + "y" * 400_000) # ~400KB -> over budget
doc = generate_task_doc(session_dir=s, phase=7)
assert doc.byte_budget_ok is False
def test_task_doc_carries_phase_header(tmp_path):
s = tmp_path / "sess"; s.mkdir()
(s / "question.md").write_text("q")
doc = generate_task_doc(session_dir=s, phase=4)
assert "fase 4" in doc.body.lower() or "fase 4" in doc.body
def test_task_doc_excludes_stale_decisions_post_rollback(tmp_path):
"""D15+D16: il task doc riflette lo stato effective, non quello stale.
Dopo rollback a F4, una sql_approved:7 stale non appare nel brief delle decisioni."""
from nsp.decisions import append_decision
from nsp.phase import current_phase
s = tmp_path / "sess"; s.mkdir()
(s / "question.md").write_text("q")
# simula: lavoro fino a F7, poi rollback a F4
append_decision(s, type="phase_approved", subject="phase:1")
append_decision(s, type="phase_approved", subject="phase:2")
append_decision(s, type="phase_approved", subject="phase:3")
append_decision(s, type="phase_approved", subject="phase:4")
append_decision(s, type="sql_approved", subject="phase:7", detail="SELECT 1")
append_decision(s, type="phase_reopened", subject="phase:4")
assert current_phase(s) == 4
doc = generate_task_doc(session_dir=s, phase=4)
# la decisione stale di fase 7 NON deve apparire nel brief
assert "sql_approved" not in doc.body
assert "phase:7" not in doc.body