feat(inbox): C2.5 — inbox hardening + SDK version guard (closes C-F7, C-N3, R-6)

- File-layer decision vocabulary (§4.2 set) with SKIP semantics — an unknown
  decision never reaches the store (C-F7, the review's run proof is the fixture)
- Fail-fast caps (max_files / max_rationale_chars) via InboxLimitError raised
  OUTSIDE the tolerant try — a cap breach is never swallowed as a skip
- R-6 id grammar (mirrors ingest _ID_RE) as a pydantic pattern on
  VerdictDocument.id AND re-checked in write_verdict, since model_copy(update=)
  bypasses model validation — traversal ids can no longer write outside the inbox
- promotion._filename_token: any sanitised id maps to a content hash — 'e/vil'
  can no longer clobber the distinct id 'evil' (restarbeid-funn 2)
- SDK pinned >=0.2.111,<0.3 + version guard test naming the sdk_client.py
  attribute premises; resolved 0.2.120, all premises re-verified against it
- sdk_client read loop bound offline with REAL SDK message types (R-4/R-5):
  text aggregation, error fail-paths, usage/cost extraction, _total_tokens
  fail-closed, non-positive budget guard
- test_sdk_isolation comment no longer claims the --system-prompt ""
  serialization the test body does not bind (honesty rule §1)

Guard-G2 assessment (guard-plan §4): the allowlist + caps + id grammar landed
here are G2's necessary part; an optional scan_output depth pass over
rationale (still a verbatim prose channel into the fold prompt, R-9) remains
relevant as a later additive session — the trigger picture is unchanged.

4 detach proofs red → restored green. Full gate: 389 passed (365→389),
ruff+format+mypy clean; golden + shared/ + runs/s10/ byte-untouched.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Kjell Tore Guttormsen 2026-07-16 20:26:41 +02:00
commit 80a2fa1a77
9 changed files with 441 additions and 35 deletions

View file

@ -13,7 +13,9 @@ from __future__ import annotations
import json
from pathlib import Path
import pytest
from _scripted import ScriptedClient, reply
from pydantic import ValidationError
from portfolio_optimiser_claude.budget import BudgetMeter
from portfolio_optimiser_claude.contracts import TerminationContract
@ -25,6 +27,7 @@ from portfolio_optimiser_claude.experience import (
mint_verdict_id,
)
from portfolio_optimiser_claude.inbox import (
InboxLimitError,
ProposalFeatures,
VerdictDocument,
load_inbox,
@ -162,6 +165,98 @@ class TestTolerantLoad:
assert [d.id for d in load_inbox(tmp_path)] == sorted([late.id, early.id])
class TestFileLayerVocabulary:
"""C2.5 (C-F7): the file layer polices the §4.2 decision set — SKIP, never raise."""
def test_an_unknown_decision_never_reaches_the_store(self, tmp_path: Path) -> None:
# Fixture = the review's RUN C-F7 proof: before C2.5 this string
# entered the store and its rationale folded into the next prompt.
doc = _document().model_copy(update={"decision": "hva-som-helst"})
(tmp_path / f"{doc.id}.json").write_text(json.dumps(doc.model_dump()), encoding="utf-8")
store = VerdictStore()
assert merge_inbox_into_store(store, tmp_path) == 0
assert len(store) == 0
def test_the_full_file_layer_vocabulary_loads(self, tmp_path: Path) -> None:
# Control: exactly the §4.2 set {approved, rejected,
# approved_with_adjustment} passes the file layer.
decisions = ("approved", "rejected", "approved_with_adjustment")
for index, decision in enumerate(decisions):
write_verdict(
tmp_path, _document(_features(codes=frozenset({f"C{index}"})), decision=decision)
)
assert {d.decision for d in load_inbox(tmp_path)} == set(decisions)
class TestInboxCaps:
"""C2.5: configurable caps fail FAST with a precise error — never a silent cut."""
def test_rationale_over_the_cap_fails_fast_with_a_precise_error(self, tmp_path: Path) -> None:
doc = _document(rationale="x" * 201)
write_verdict(tmp_path, doc)
with pytest.raises(InboxLimitError) as err:
load_inbox(tmp_path, max_rationale_chars=200)
message = str(err.value)
assert doc.id in message
assert "201" in message
assert "200" in message
def test_rationale_at_the_cap_loads(self, tmp_path: Path) -> None:
write_verdict(tmp_path, _document(rationale="x" * 200))
assert len(load_inbox(tmp_path, max_rationale_chars=200)) == 1
def test_more_files_than_the_cap_fails_fast(self, tmp_path: Path) -> None:
for index in range(3):
write_verdict(tmp_path, _document(_features(codes=frozenset({f"C{index}"}))))
with pytest.raises(InboxLimitError) as err:
load_inbox(tmp_path, max_files=2)
message = str(err.value)
assert "3" in message
assert "2" in message
def test_merge_threads_the_caps_through(self, tmp_path: Path) -> None:
write_verdict(tmp_path, _document(rationale="x" * 201))
with pytest.raises(InboxLimitError):
merge_inbox_into_store(VerdictStore(), tmp_path, max_rationale_chars=200)
class TestIdGrammar:
"""C2.5 (R-6): the id grammar is policed BEFORE ``write_verdict`` touches disk."""
def test_a_traversal_id_is_rejected_at_construction(self) -> None:
# The security agent's RUN R-6 proof: '../../escaped' used to
# construct fine and write outside the inbox.
with pytest.raises(ValidationError):
VerdictDocument(
id="../../escaped",
decision="approved",
rationale=RATIONALE,
proposal_features=ProposalFeatures(
affected_codes=["E01"],
measure_type="led-retrofit",
claimed_saving_nok=25000.0,
description="surface text",
),
)
def test_write_verdict_refuses_an_escaping_id_writing_nothing(self, tmp_path: Path) -> None:
# model_copy(update=...) bypasses model validation — the write seam
# must fail closed on its own (detach the grammar here → a file lands
# OUTSIDE the inbox → red).
inbox = tmp_path / "inbox"
doc = _document().model_copy(update={"id": "../../escaped"})
with pytest.raises(ValueError):
write_verdict(inbox, doc)
assert not (tmp_path / "escaped.json").exists()
assert not inbox.exists()
def test_a_loaded_file_with_an_escaping_id_is_skipped(self, tmp_path: Path) -> None:
# Tolerant load stays tolerant: a hostile id is SKIPPED, never raised.
doc = _document().model_copy(update={"id": "../../escaped"})
(tmp_path / "escaped.json").write_text(json.dumps(doc.model_dump()), encoding="utf-8")
assert load_inbox(tmp_path) == []
class TestMergeIntoStore:
"""§5: merge, never replace — first-write-wins per id, idempotent, read-only."""