Dodaios is an open-source agent OS for your desktop: it runs your coding agents, browser and apps by voice or chat, on your own machine. Formerly Personal Jarvis.
Python agent-orchestrationai-agentai-coding-assistantclaude-codecoding-agentcomputer-usedesktop-applocal-firstmcppersonal-jarvisvoice-assistantvoice-coding
"""Unit tests for ``dodaios.memory.wiki.prompt`` (Instance D)."""
from __future__ import annotations
from pathlib import Path
from types import SimpleNamespace
from typing import Any
import pytest
from dodaios.memory.wiki.prompt import (
build_consolidator_prompt,
build_system_prompt,
build_user_prompt,
compute_vault_summary,
resolve_user_entity_slug,
select_top_slugs,
)
# ---------------------------------------------------------------------
# Small in-memory VaultIndex fake — only ``pages_by_type`` is used.
# ---------------------------------------------------------------------
class _Page:
"""Minimal duck-typed page with a slug + page_type."""
def __init__(self, slug: str, page_type: str) -> None:
self.slug = slug
self.page_type = page_type
class _FakeVault:
"""In-memory ``VaultIndex`` substitute for prompt tests."""
def __init__(self, pages_by_type: dict[str, list[_Page]] | None = None) -> None:
self._pages = pages_by_type or {}
def pages_by_type(self, page_type: str) -> list[Any]:
return list(self._pages.get(page_type, []))
# ---------------------------------------------------------------------
# select_top_slugs
# ---------------------------------------------------------------------
def test_select_top_slugs_empty_source_returns_empty() -> None:
"""A blank source contributes no tokens, so the shortlist is empty."""
result = select_top_slugs("", ["ruben-luetke", "awareness-layer"])
assert result == []
def test_select_top_slugs_zero_overlap_returns_empty() -> None:
"""When no candidate slug shares a token with the source, drop them all."""
result = select_top_slugs("Phase B1 wiki curator", ["unrelated-thing"])
assert result == []
def test_select_top_slugs_ranks_by_overlap_then_alpha() -> None:
"""Higher overlap wins; ties break alphabetically for deterministic output."""
source = "ruben pushes the awareness layer for the jarvis-agent bridge"
candidates = [
"ruben-luetke",
"awareness-layer",
"jarvis-agent-bridge",
"phase-6",
"kontrollierer",
]
ranked = select_top_slugs(source, candidates)
assert "awareness-layer" in ranked
assert "jarvis-agent-bridge" in ranked
assert "ruben-luetke" in ranked
assert "kontrollierer" not in ranked
def test_select_top_slugs_respects_limit() -> None:
"""The hard cap is honoured even when many candidates score."""
source = " ".join(f"slug{i}" for i in range(30))
candidates = [f"slug{i}" for i in range(30)]
result = select_top_slugs(source, candidates, limit=10)
assert len(result) == 10
def test_select_top_slugs_ignores_stopwords() -> None:
"""Common DE/EN function words don't dominate the ranking."""
source = "der und die das the and the jarvis-agent bridge" # i18n-allow: German/English stopword mix, content under test
candidates = ["der-thing", "jarvis-agent-bridge"]
result = select_top_slugs(source, candidates)
assert result and result[0] == "jarvis-agent-bridge"
# ---------------------------------------------------------------------
# compute_vault_summary
# ---------------------------------------------------------------------
def test_compute_vault_summary_empty_vault() -> None:
"""An empty vault yields zero counts and an empty recent-log list."""
vault = _FakeVault()
summary = compute_vault_summary(vault)
assert summary["counts"] == {
"entity": 0,
"concept": 0,
"project": 0,
"session": 0,
}
assert all(slugs == [] for slugs in summary["latest"].values())
assert summary["recent_log"] == []
def test_compute_vault_summary_counts_and_samples(tmp_path: Path) -> None:
"""Page counts match; ``latest`` is alphabetical and capped at 5 slugs."""
vault = _FakeVault({
"entity": [_Page(f"e-{i:02d}", "entity") for i in range(7)],
"concept": [_Page("awareness-layer", "concept")],
})
summary = compute_vault_summary(vault)
assert summary["counts"]["entity"] == 7
assert summary["counts"]["concept"] == 1
assert summary["latest"]["entity"] == ["e-00", "e-01", "e-02", "e-03", "e-04"]
assert summary["latest"]["concept"] == ["awareness-layer"]
def test_compute_vault_summary_reads_recent_log_entries(tmp_path: Path) -> None:
"""Three most recent ``## [...]`` headings are returned in chronological order."""
log = tmp_path / "log.md"
log.write_text(
"# Wiki Log\n\n"
"## [2026-05-11 18:00] create | first\n\n- a\n\n"
"## [2026-05-11 18:30] update | second\n\n- b\n\n"
"## [2026-05-11 19:00] update | third\n\n- c\n\n"
"## [2026-05-11 19:30] update | fourth\n\n- d\n",
encoding="utf-8",
)
summary = compute_vault_summary(_FakeVault(), log_path=log)
assert len(summary["recent_log"]) == 3
assert summary["recent_log"][-1].startswith("[2026-05-11 19:30]")
def test_compute_vault_summary_missing_log_is_silent(tmp_path: Path) -> None:
"""A non-existent log path degrades to an empty list, no raise."""
summary = compute_vault_summary(_FakeVault(), log_path=tmp_path / "nope.md")
assert summary["recent_log"] == []
def test_compute_vault_summary_resilient_to_vault_errors() -> None:
"""A vault that raises on ``pages_by_type`` still yields zero counts."""
class _BoomVault:
def pages_by_type(self, page_type: str) -> list[Any]: # noqa: ARG002
raise RuntimeError("vault is on fire")
summary = compute_vault_summary(_BoomVault())
assert summary["counts"]["entity"] == 0
# ---------------------------------------------------------------------
# build_system_prompt
# ---------------------------------------------------------------------
def test_build_system_prompt_no_longer_embeds_schema_md() -> None:
"""``schema.md`` is vault documentation; the prompt carries the contract itself."""
prompt = build_system_prompt(None)
assert "# Wiki Schema" not in prompt
assert "Pages: frontmatter type, slug" in prompt
def test_build_system_prompt_appends_output_contract() -> None:
"""The JSON output contract is always part of the system prompt."""
prompt = build_system_prompt(vault_summary=None)
assert "Output Contract" in prompt
assert '"operation"' in prompt
assert "create" in prompt and "update" in prompt
assert "[]" in prompt
def test_build_system_prompt_includes_vault_summary_when_provided() -> None:
"""``compute_vault_summary``'s output is rendered into the prompt."""
vault_summary = {
"counts": {"entity": 2, "concept": 1, "project": 0, "session": 0},
"latest": {
"entity": ["ruben-luetke", "personal-jarvis"],
"concept": ["awareness-layer"],
"project": [],
"session": [],
},
"recent_log": ["[2026-05-11 18:00] create | seed"],
}
prompt = build_system_prompt(vault_summary=vault_summary)
assert "Entities: 2" in prompt
assert "ruben-luetke" in prompt
assert "Concepts: 1" in prompt
assert "[2026-05-11 18:00] create | seed" in prompt
def test_build_system_prompt_handles_no_summary() -> None:
"""Passing ``None`` skips the snapshot section but keeps the contract."""
prompt = build_system_prompt(vault_summary=None)
assert "Current Vault Snapshot" not in prompt
assert "Output Contract" in prompt
# ---------------------------------------------------------------------
# build_user_prompt
# ---------------------------------------------------------------------
def test_build_user_prompt_wraps_source_verbatim() -> None:
"""Source content is rendered between explicit BEGIN/END markers."""
prompt = build_user_prompt(
"BrainTurnCompleted 2026-05-11 19:42",
"Ruben fixed BUG-019 in fix/bug-019-tts-silent.",
top_slugs=["ruben-luetke"],
)
assert "Ruben fixed BUG-019 in fix/bug-019-tts-silent." in prompt
assert "----- BEGIN SOURCE -----" in prompt
assert "----- END SOURCE -----" in prompt
assert "BrainTurnCompleted 2026-05-11 19:42" in prompt
def test_build_user_prompt_includes_top_slugs() -> None:
"""The keyword-overlap shortlist is rendered as bullet hints."""
prompt = build_user_prompt(
"source-label", "some content", top_slugs=["ruben-luetke", "jarvis-agent-bridge"],
)
assert "- ruben-luetke" in prompt
assert "- jarvis-agent-bridge" in prompt
def test_build_user_prompt_handles_no_top_slugs() -> None:
"""The "no overlap" hint is emitted when the shortlist is empty."""
prompt = build_user_prompt("source-label", "some content", top_slugs=[])
assert "no overlap detected" in prompt
def test_build_user_prompt_handles_none_top_slugs() -> None:
"""``None`` is treated like an empty list (no crash on missing list)."""
prompt = build_user_prompt("source-label", "content")
assert "no overlap detected" in prompt
def test_build_user_prompt_demands_json_at_end() -> None:
"""The user message ends with the explicit JSON-array directive."""
prompt = build_user_prompt("label", "content", top_slugs=[])
assert prompt.rstrip().endswith("Return the JSON array now.")
# ---------------------------------------------------------------------
# Salience filter (B8): the system prompt must teach the LLM which
# inputs to drop ([]) and which to persist. We can't unit-test the LLM
# itself, but we can pin the prompt content so a future refactor
# cannot quietly delete the filter.
# ---------------------------------------------------------------------
def test_system_prompt_states_the_applied_checks() -> None:
"""The deterministic checks the model cannot know are stated as facts."""
from dodaios.memory.wiki.curator_llm import _MAX_UPDATES_PER_INGEST
prompt = build_system_prompt(vault_summary=None)
assert f"At most {_MAX_UPDATES_PER_INGEST} updates apply" in prompt
assert "becomes plain text" in prompt
assert "<!-- contact-mirror:start -->" in prompt
def test_system_prompt_has_no_maintainer_specific_user_target() -> None:
"""Generic curator guidance never bakes in one maintainer's profile slug."""
prompt = build_system_prompt(vault_summary=None)
assert "entities/ruben.md" not in prompt.lower()
assert 'exact subject slug is ["user"]' in prompt
# ---------------------------------------------------------------------
# Stage-2 user-entity binding
# ---------------------------------------------------------------------
@pytest.mark.parametrize(
("configured", "expected"),
[
("", "user"),
(None, "user"),
("../../private", "user"),
("Owner-Profile", "owner-profile"),
],
)
def test_resolve_user_entity_slug_is_safe_and_neutral(
configured: object,
expected: str,
) -> None:
assert resolve_user_entity_slug(configured) == expected
@pytest.mark.parametrize(
("configured", "expected"),
[("", "user"), ("owner-profile", "owner-profile")],
)
def test_consolidator_prompt_binds_exact_user_subject_and_page(
configured: str,
expected: str,
) -> None:
candidate = SimpleNamespace(
id=1,
fact="The speaker prefers dark mode.",
kind="preference",
subjects=(expected,),
)
system, user = build_consolidator_prompt(
[candidate],
{},
user_entity_slug=configured,
)
assert f'subject slug ["{expected}"]' in user
assert f"profile page entities/{expected}.md" in user
assert "never infer a name" in user
assert "ruben" not in f"{system}\n{user}".lower()
def test_consolidator_prompt_rechecks_question_derived_user_claims() -> None:
candidate = SimpleNamespace(
id=7,
fact="The user is interested in Vitamin D.",
kind="preference",
subjects=("user",),
evidence_turn_id="vitamin-turn",
evidence_excerpt=(
"Evidence user turn [vitamin-turn]: "
"What are the benefits of Vitamin D?"
),
)
system, user = build_consolidator_prompt([candidate], {})
assert "asserts or confirms it" in system
assert "Vitamin D" not in system
assert "What are the benefits of Vitamin D?" in user
# Rows without basis metadata render the explicit/3 defaults.
assert "basis=explicit salience=3" in user
def test_consolidator_prompt_renders_the_validator_flags_per_candidate() -> None:
place = SimpleNamespace(
id=9,
fact="The user lives in San Francisco.",
kind="place",
subjects=("user", "san-francisco"),
)
plan = SimpleNamespace(
id=8,
fact="The user plans to travel to San Francisco tomorrow.",
kind="plan",
subjects=("user",),
)
system, user = build_consolidator_prompt(
[place, plan],
{},
row_flags={
9: "persist_requested=no required_page=entities/san-francisco.md",
8: "persist_requested=yes required_page=-",
},
)
assert (
"candidate_id=9 kind=place subjects=[user, san-francisco] basis=explicit "
"salience=3 persist_requested=no required_page=entities/san-francisco.md"
) in user
assert "persist_requested=yes required_page=-" in user
assert "required_page=<path>" in system
assert "unsupported by user evidence" in system
def test_consolidator_prompt_without_flags_keeps_the_plain_candidate_line() -> None:
row = SimpleNamespace(id=3, fact="f", kind="preference", subjects=("user",))
_system, user = build_consolidator_prompt([row], {})
assert "persist_requested" not in user
assert "basis=explicit salience=3\n" in user
# ---------------------------------------------------------------------
# Integration-style assertion: round-trip through everything once.
# ---------------------------------------------------------------------
def test_build_full_prompt_roundtrip_smoke() -> None:
"""One full pipeline: vault → summary → system prompt + user prompt."""
vault = _FakeVault({
"entity": [_Page("ruben-luetke", "entity"), _Page("personal-jarvis", "entity")],
"concept": [_Page("awareness-layer", "concept")],
})
summary = compute_vault_summary(vault)
system_prompt = build_system_prompt(summary)
top_slugs = select_top_slugs("Ruben pushed the awareness layer", [
"ruben-luetke", "awareness-layer", "personal-jarvis",
])
user_prompt = build_user_prompt(
"BrainTurnCompleted", "Ruben pushed the awareness layer", top_slugs,
)
assert "Entities: 2" in system_prompt
assert "Concepts: 1" in system_prompt
assert "ruben-luetke" in user_prompt
assert "awareness-layer" in user_prompt