text
stringlengths
185
73.3k
repo
stringlengths
7
100
path
stringlengths
4
146
language
stringclasses
7 values
hash
stringlengths
16
16
score
float64
7
8.5
stars
int64
0
237k
"""honest geometric ray-tracer primitives · read-only.""" from __future__ import annotations import numpy as np def thin_lens_refract( points: np.ndarray, # (N, 3) at lens plane (z=0 slice) directions: np.ndarray, # (N, 3) ray direction before lens z_local: float, # not used (thin-lens at plane) ...
chunxiaoxx/nautilus-compass
_scratch_parking/optics_poc_lens_pair/Optics/lens_pair_aberration_v1_001/verification/_core.py
.py
207af6572efdf360
7.45
7
"""Honest geometric ray-tracing verifier · read-only. Imports `align_lens_pair` from `baseline/init.py`, ray-traces through both lenses, measures RMS spot radius on the target focal plane. """ from __future__ import annotations import argparse import json import importlib.util import sys from pathlib import Path im...
chunxiaoxx/nautilus-compass
_scratch_parking/optics_poc_lens_pair/Optics/lens_pair_aberration_v1_001/verification/evaluate.py
.py
acc91c611ad3eb67
7.45
7
# -*- coding: utf-8 -*- """86 题正解一致性自检(抽样版)。 双门(对每题,git worktree 隔离): 门1 starter 必败:worktree@buggy_commit + 测试@fix_commit → pytest 必须 FAIL 门2 正解必过:再把 starter_path 换成 @reference_fix_commit 版 → pytest 必须 PASS """ import json import subprocess import sys import tempfile from pathlib import Path REPOS = { "nautil...
chunxiaoxx/nautilus-compass
_v5_proof_deposit/g2b1/_g2b1_consistency_check.py
.py
5f9553455d52813c
7.45
7
"""compass ALE-Bench scorer · wraps ale_bench public_eval into an rsi_two_arm eval_fn. 齿轮⑤/③ 旁支 → 胜负手燃料线(ALE 连续分·非饱和)。eval_fn 跑在 T4 ALE env (需 ale_bench 1.5.0 + cpp23 判题镜像 · /workdir 须 1777 · 见 docs)。 设计 = effect 注入(start_fn)· 与 V5 `ale_fuel_batch.py` 的 strong_fn/doubao_fn 注入 风格一致(anchor#5 不另造)。单测注 fake session 无需 li...
chunxiaoxx/nautilus-compass
ale_bench/ale_eval.py
.py
5041db05e299081f
7.45
7
#!/usr/bin/env python3 """nautilus-compass KPI audit · 看真实使用率 · 1 周后跑. Adds (2026-05-30 · E.fix-2): `act_on_rate()` function reading drift_mitigation_log.jsonl to compute act-on rate (closes 5/27 finding's open loop · detection side ~25k events · ack side previously 0). """ import json import sys from collections impo...
chunxiaoxx/nautilus-compass
audit_kpi.py
.py
8a8afac2ec28c34b
7.45
7
#!/usr/bin/env python3 """v2.0.0 · #2 · self-evolve CLI · fallback for SessionEnd-hook missed runs. Triggers L1 collapse + entity link scan for one project or all projects. Designed for cron / systemd timer · every 1h is a reasonable backstop. Usage: python bin/cli_self_evolve.py --project <slug> # specific pro...
chunxiaoxx/nautilus-compass
bin/cli_self_evolve.py
.py
e354db28e01eaa0f
7.45
7
"""v1.6.0 · 用户 #1 真融合 · 把本地 session_*.md push 到 cloud /v1/v14/ingest_obs V7-N5 phase 2 P0: 本地 stop_hook 已写 session_*.md + drift sidecar · 但 cloud 不知道(local 不 sync cloud)。 V5/Kairos/Souls Fusion 的 cross-dialog recall 看不到我(claude-code-compass-dialog)写的。 修法: 每次 stop_hook 新写一份 session_*.md 后 · SSH inline POST 到 cloud:877...
chunxiaoxx/nautilus-compass
cloud_ingest.py
.py
cdaa17863c907b3c
7.45
7
"""compass fleet 记忆胶囊客户端(cross-agent 集体学习)· 2026-06-17 · compass 交付给 V5/platform 把"飞轮 agent learning 写回 + claim 前 recall"降到两个 import 即用的函数,V5 无需自己写 HTTP。 - W1: write_learning(agent_id, task_family, reason) 解完 reward=1.0 后调 - W2: compass_recall_pits(task_family) -> [{item_id,reason}] claim/产出前调·直接喂 V5 fd...
chunxiaoxx/nautilus-compass
compass_fleet_memory.py
.py
a9919039ee27673c
7.45
7
"""compass v1.0 · RAID-2 (writer-reviewer) write path · #7 fusion. For org / enterprise plan: every observation must pass a reviewer-agent audit before being written to memory. The reviewer is compass's own anchor + LLM judge · NOT the writer agent itself (defeats purpose). Workflow: 1. Writer submits obs → POST /v...
chunxiaoxx/nautilus-compass
compass_raid.py
.py
95b0ea575f3117a9
7.45
7
#!/usr/bin/env python3 """compass verify · v1.0 tamper-evidence CLI. Walks ~/.claude/projects/<encoded>/memory/ directories, verifies the Merkle hash chain recorded in .chain.json, reports tampered / missing files. Exit code: 0 · all verified chains are valid 1 · any tampered or missing file detected, or chain fi...
chunxiaoxx/nautilus-compass
compass_verify.py
.py
f2c2f982cb6d5e99
7.45
7
"""齿轮⑤ 深 consolidation:36k cycle-auto 日记碎片 → 挑信号做胶囊 + 噪声归档(可逆)。 实测(2026-06-22):~/.claude/projects 下 36032 个 cycle-*-auto 目录(93%)·95% 单文件· 内容 = V5 daemon 每 cycle 的 think:/learn:/evolve: 独白·抽样 learn: 字段全空 = 噪声非教训。 把它们整体做成胶囊会淹没真信号(反精炼)→ 正解:挑 learn: 非空的少数蒸馏·其余归档。 设计 = 纯解析/分类(TDD)+ 注入 I/O(walk/write/archive)。归档 = mv 可逆·非删...
chunxiaoxx/nautilus-compass
cycle_consolidator.py
.py
1ff9b223f9695389
7.45
7
"""compass v0.9 · platform_anchors layered loader · #6 fusion. 合并 3 层 anchors: 1. anchors_platform_base.json · 平台通用 (15 pos + 25 neg) 2. anchors_<domain>.json · 行业 (finance/legal/medical/vc/zenmind) 3. anchors.json · tenant/user 个人补充 输出: list[positive], list[negative] · 喂给 daemon BGE 索引 调...
chunxiaoxx/nautilus-compass
daemon_anchor_loader.py
.py
6397c36a375110e8
7.45
7
#!/usr/bin/env python3 """V5 Memory Plugin · 深度边界测试 · 跑慢但真. Test 1: hook 模式 (无 BGE) 速度真 < 1s Test 2: BGE cold cache · 真冷启 Test 3: BGE warm cache · 应 < cold (即便 model load 是 fixed) Test 4: anchors.json mtime 变 → 重 embed Test 5: memory file mtime 变 → 部分重 embed (不是全部) Test 6: 跨项目 (cwd 切到非 nautilus-core) Test 7: 特殊 prompt...
chunxiaoxx/nautilus-compass
deeptest.py
.py
9a261eaf0bd515e0
7.95
7
"""E.fix-1 (2026-05-30) · drift alert ack writer + reader. Background (5/27 finding · session_20260527_drift_loop_open_tuneout): detection side wrote 25k+ events to `.cache/drift_mitigation_log.jsonl` (recall.py:159 writer) with stable `alert_id` per fire · but ZERO ack records were ever written → act-on rate uncomput...
chunxiaoxx/nautilus-compass
drift/act_log.py
.py
67a2775da7639bcc
7.45
7
"""H.1 (2026-05-30) · stop_hook auto-detect drift alert acks in session text. Closes the 5/27 drift loop open finding's last unmeasured side: agent self-acknowledgement of drift alerts (vs the user-CLI-only path via E.fix-3). Algorithm (rule-based · no LLM · stop_hook policy): 1. Find every `a-XXXXXXXX` literal (8 he...
chunxiaoxx/nautilus-compass
drift/auto_ack.py
.py
756d791c822f94f2
7.45
7
"""Desktop notifications. `notify(state, project, detail)` is the single entry point. It respects the `@ccm-notify` tmux option (off / permit / completed / permit,completed / all) and dispatches via the best available backend: - macOS, terminal-notifier installed → `terminal-notifier -group ccm-<project>` so a ...
yohasebe/tmux-ccm
lib/ccm_notify.py
.py
46205ecdf27bb089
7.59
14
"""Declarative state-detection rule table + evaluator. This module owns the **legacy** detection path: the priority-ordered `DETECTION_RULES` table and its supporting types (`DetectionContext`, `Rule`, `Action`, `USE_RAW`). It runs as the safety net when `ccm_activity.derive_state_from_events` returns None (empty even...
yohasebe/tmux-ccm
lib/ccm_rules.py
.py
1fe807f04fcd811f
7.59
14
"""Project-state persistence: save / load / list / delete snapshots. A snapshot is a JSON manifest at `$CCM_SNAPSHOT_DIR/<name>.json` recording every ccm-tagged tmux window: `(name, dir, auto_start_claude)`. The `_autosave` snapshot written by `cmd_snapshot_save("_autosave", quiet=True)` is the restore point for `ccm ...
yohasebe/tmux-ccm
lib/ccm_snapshot.py
.py
78e5cfdeee82cd7a
7.59
14
"""Shared test infrastructure: lib path setup, autouse fixtures, and helpers used across the per-module test files.""" import json import os import sys import time from datetime import datetime, timezone import pytest # Make lib/ importable so each test file can `import ccm_core` etc. sys.path.insert(0, os.path.join...
yohasebe/tmux-ccm
tests/conftest.py
.py
ef7fb735825d413c
8.09
14
"""Tests for ccm_agentview — read-only access to Claude Code's per-user agent-view daemon (roster.json + jobs/<short>/state.json).""" import json import os import sys import pytest sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "lib")) import ccm_agentview @pytest.fixture def fake_claude_home(tmp...
yohasebe/tmux-ccm
tests/test_agentview.py
.py
1742fb651f76a5bd
8.09
14
"""Tests for ccm_core.py — state detection, helpers, and batch tmux commands.""" import json import os import re import sys import time from datetime import datetime, timezone from unittest.mock import patch, MagicMock, call import pytest # Add lib to path sys.path.insert(0, os.path.join(os.path.dirname(__file__), "...
yohasebe/tmux-ccm
tests/test_ccm_core.py
.py
fc07c85b447ba8f8
7.09
14
"""Property-based tests for `derive_state_from_events`. Where the parametric tests in `test_ccm_core.py::TestDeriveStateFromEvents` exercise specific hook sequences, this file verifies invariants that must hold across the entire input space: - never raises (any input → either a valid state or None) - returns a va...
yohasebe/tmux-ccm
tests/test_derive_invariants_pbt.py
.py
92974460375ab5ce
7.09
14
"""Tests for the CCM_IGNORE feature: hiding a pane/session from ccm. Covers the predicate (`_pane_is_ignored`), the count helper, the raw pane-aggregation skip, and the `ccm ignore` / `ccm unignore` command markers. The bash side (hook early-exit + pane stamp) is covered in tests/test_event_log.bats. """ import os fr...
yohasebe/tmux-ccm
tests/test_ignore.py
.py
541f6ecd6d831d11
8.09
14
"""Tests for ccm_pane_state. Auto-split from test_ccm_core.py. Shared fixtures + helpers (write_jsonl, make_ps_lines, real_activity_record, system_record, iso_ts) live in conftest.py; import them here when used. """ import json import os import re import sys import time from datetime import datetime, timezone from un...
yohasebe/tmux-ccm
tests/test_pane_state.py
.py
050ef930b377ddaa
7.09
14
"""Tests for ccm_rules. Auto-split from test_ccm_core.py. Shared fixtures + helpers (write_jsonl, make_ps_lines, real_activity_record, system_record, iso_ts) live in conftest.py; import them here when used. """ import json import os import re import sys import time from datetime import datetime, timezone from unittes...
yohasebe/tmux-ccm
tests/test_rules.py
.py
6c7d498f1b0f7cec
8.09
14
"""Tests for `ccm sidekick-send` — delivery to the external-agent sidekick pane of the caller's own window. The command's safety story is identity, not state: ccm tracks no state for a non-Claude pane, so the target is found from tmux metadata (foreground command + working directory), every ambiguity refuses, and deli...
yohasebe/tmux-ccm
tests/test_sidekick_send.py
.py
e17103ab465fd952
8.09
14
"""Tests for ccm_signals. Auto-split from test_ccm_core.py. Shared fixtures + helpers (write_jsonl, make_ps_lines, real_activity_record, system_record, iso_ts) live in conftest.py; import them here when used. """ import json import os import re import sys import time from datetime import datetime, timezone from unitt...
yohasebe/tmux-ccm
tests/test_signals.py
.py
e6c3a52fde0a92cf
7.09
14
"""Regression tests for the "silent exception class" of bug. The autosave NameError that ran for 38 hours, the notify dedup keying mismatch, and the multi-line ❯ false-BUSY were all silent because their failure paths were caught by broad `except Exception:` clauses or by `log_caught_exception`. Tests that mock the cal...
yohasebe/tmux-ccm
tests/test_silent_exceptions.py
.py
34f34d1ac679dcfd
8.09
14
"""Trace-replay regression corpus. Replays REAL captured event logs (hooks/<sessionId>.events.jsonl files lifted verbatim from production incidents) through `derive_state_from_events` and asserts the expected state at probe points along the timeline. Why this exists: every detection bug ccm has had shares one shape —...
yohasebe/tmux-ccm
tests/test_trace_replay.py
.py
2361210d2282cb2f
8.09
14
import json import os from typing import Annotated, TypedDict from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage from langchain_core.tools import tool from langchain_openai import ChatOpenAI from langgraph.graph import END, StateGraph from langgraph.graph.message import add_message...
huaweicloud/agentarts-sdk-python
examples/agent_tools/integrate_tools.py
.py
1349b91a9fb1d8ee
7.6
15
"""LangChain Integration Example - Agent with tools using AgentArts SDK""" import os from langchain.agents import AgentExecutor, create_tool_calling_agent from langchain_core.prompts import ChatPromptTemplate from langchain_core.tools import tool from langchain_openai import ChatOpenAI from agentarts.sdk import Agen...
huaweicloud/agentarts-sdk-python
examples/integration/langchain/langchain_agent.py
.py
06886b710b77ce42
7.6
15
""" Navigation Agent Demo - Memory Recall Tool Exposes a single LangChain tool, recall_memory, that performs semantic search over the AgentArts Memory service to retrieve long-term memories (user preferences, historical facts, etc.). This is the ON-DEMAND component of the hybrid recall architecture. The AUTO-INJECTIO...
huaweicloud/agentarts-sdk-python
examples/navigation_langgraph_memory/memory_tools.py
.py
84cb2a824718f293
7.6
15
"""Navigation Agent Demo - Message Utilities Shared helpers for extracting message content and fetching session history. Used by both CLI (nav_agent.py) and TUI (tui_app.py) to avoid duplication. """ from typing import List, Tuple import config # noqa: F401 from config import API_KEY, SPACE_ID, VERIFY_SSL def ext...
huaweicloud/agentarts-sdk-python
examples/navigation_langgraph_memory/message_utils.py
.py
a7a3ec7a637337bf
7.6
15
""" Navigation Agent Demo - LangGraph Agent with AgentArts Memory A local interactive navigation assistant that can: - Search for POIs (gas stations, restaurants, parking, etc.) - Plan routes (driving/walking/riding/transit) - Generate map visualization links - Recall long-term user preferences and history via...
huaweicloud/agentarts-sdk-python
examples/navigation_langgraph_memory/nav_agent.py
.py
e9190f03f81198e0
7.6
15
""" Navigation Agent Demo - Session Manager Manages multiple conversation sessions locally. Each session maps to an AgentArts Memory session (via session_id). Session metadata is persisted to sessions.json so sessions survive restarts. Usage: from session_manager import select_session_interactive session_id =...
huaweicloud/agentarts-sdk-python
examples/navigation_langgraph_memory/session_manager.py
.py
14f6215b6aca67a2
7.6
15
""" Navigation Agent Demo - Memory Space Setup (run once) Creates a new AgentArts Memory space with all four builtin extraction strategies. Prints the Space ID and API Key that must be exported as environment variables before running nav_agent.py. Sessions are NOT created here — session_manager.py creates a new Agent...
huaweicloud/agentarts-sdk-python
examples/navigation_langgraph_memory/setup_memory.py
.py
7997ce88191f637d
7.6
15
"""Navigation Agent TUI application using Textual framework.""" import asyncio import io import threading from typing import Optional from textual.app import App, ComposeResult from textual.containers import Container from textual.screen import ModalScreen from textual.widgets import Button, Footer, Header, Input, Lab...
huaweicloud/agentarts-sdk-python
examples/navigation_langgraph_memory/tui_app.py
.py
f0d83f76c31231da
7.6
15
"""Type aliases for improved developer experience. This module provides user-friendly type aliases for types used throughout the SDK. These aliases make injected types discoverable without reading source code. """ import sys from typing import TypeAlias if sys.version_info >= (3, 11): from enum import StrEnum el...
huaweicloud/agentarts-sdk-python
src/agentarts/sdk/identity/types.py
.py
72e5c16bb5ff7b35
7.6
15
""" Configuration for AgentArts LangGraph Integration Provides configuration classes for checkpoint saver and other LangGraph integration components. """ from __future__ import annotations from typing import Any from pydantic import BaseModel try: from langchain_core.runnables import RunnableConfig LANGGR...
huaweicloud/agentarts-sdk-python
src/agentarts/sdk/integration/langgraph/config.py
.py
26294328431ae312
7.6
15
""" Agent Memory SDK - AsyncMemoryClient Async version of MemoryClient for non-blocking operations. All request parameters and response handling identical to MemoryClient. Control plane operations remain synchronous (low-frequency operations). """ import threading from typing import Any from agentarts.sdk.utils.con...
huaweicloud/agentarts-sdk-python
src/agentarts/sdk/memory/async_client.py
.py
f0c8d41aabc99208
7.6
15
""" Agent Memory SDK - Async MemorySession Async version of MemorySession for non-blocking operations. All request parameters and response handling identical to MemorySession. """ import logging from .inner.config import ( MemoryInfo, MemoryListFilter, MemoryListResponse, MemorySearchFilter, Mem...
huaweicloud/agentarts-sdk-python
src/agentarts/sdk/memory/async_session.py
.py
9a932e856cbec1d9
7.6
15
""" AgentArts Runtime Models Defines constants, enumerations, and data models used by the runtime application. """ from __future__ import annotations from enum import Enum SESSION_HEADER = "x-hw-agentarts-session-id" ACCESS_TOKEN_HEADER = "X-HW-AgentGateway-Workload-Access-Token" USER_ID_HEADER = "X-HW-AgentGateway...
huaweicloud/agentarts-sdk-python
src/agentarts/sdk/runtime/model.py
.py
1534616c51f450f6
7.6
15
""" Base HTTP Client Module Provides a base HTTP client for API calls. Other service implementations should inherit from BaseHTTPClient to make HTTP requests. The client automatically detects streaming responses based on the response ``Content-Type`` header: - **Regular** (JSON / text): ``data`` is a parsed ``dict``...
huaweicloud/agentarts-sdk-python
src/agentarts/sdk/service/http_client.py
.py
0be20c86c71f83d0
7.6
15
# -*- coding: utf-8 -*- """Numbers the engine reads wrongly, rewritten before it sees them. Leopard's front end is good at numbers up to six digits and gives up after that. Measured on Alex, what the engine does with a bare digit string: 1234 "twelve thirty four" -- read as a year 12345 "...
tgeczy/panthera-speech
panthera/addon/synthDrivers/_panthera/pantheranumbers.py
.py
dfab245639eb36a0
7.59
14
# -*- coding: utf-8 -*- """Enough of NVDA to import and drive the synthesizers outside NVDA. The fakes are installed once, here, for both generations. They used to be a copy each, and the copies had drifted: only Leopard's fake `DriverSetting` recorded its arguments, which is the whole of what `test_settings_panel.py...
tgeczy/panthera-speech
panthera/tests/conftest.py
.py
6bd1be0e403b4dc4
8.09
14
# -*- coding: utf-8 -*- """Leopard's engine fixtures. Scoped to this directory on purpose. Tiger's conftest calls its fixtures exactly the same things, and pytest keeps them apart because a conftest only reaches its own folder and below. The NVDA fakes both generations share are one level up. """ import os import pyt...
tgeczy/panthera-speech
panthera/tests/leopard/conftest.py
.py
4e1d59aaf25ceac0
8.09
14
"""A breath needs a sentence boundary inside one utterance. Measured on Alex: N sentences handed over together give N-1 breaths, at the boundaries and nowhere else. A breath is 369-399 ms of turbulent noise -- rms 76-123 against a peak of ~680, around 3000 zero crossings a second -- which is what tells it apart from ...
tgeczy/panthera-speech
panthera/tests/leopard/test_breath.py
.py
b21ee2478ab91d05
8.09
14
# -*- coding: utf-8 -*- """Arrowing by letter must not restart the engine on every keystroke. Reported by Timothy Wynn, who could hear it happening: "Interrupt the voice close to the start of the utterance, e.g., navigating rapidly by letter. You will hear that the executable runs again." `cancel()` used to...
tgeczy/panthera-speech
panthera/tests/leopard/test_leopard_interrupt_cost.py
.py
d455c24f751fc3f0
8.09
14
"""A single very long token must render audio, not one frame of silence. Reported by Brandon (@serrebidev) as issue #4: **Alex returned one frame and a status of 0 -- indistinguishable from success -- for an unbroken token of about 370 characters or more**, and was intermittently silent below that, one run in five at ...
tgeczy/panthera-speech
panthera/tests/leopard/test_long_tokens.py
.py
4fc05f808c97d935
8.09
14
# -*- coding: utf-8 -*- """The engine's sentence pause, restored between continuous-reading chunks. Inside one utterance the engine composes a pause between sentences -- about half a second at 180 wpm -- and a chunk ends with zero trailing frames, so during say-all every chunk boundary slammed shut (panthera-speech#10...
tgeczy/panthera-speech
panthera/tests/leopard/test_sentence_pause.py
.py
cd5427061fb05262
8.09
14
# -*- coding: utf-8 -*- """The settings panel, checked the way it used to be checked by eye. Two labels claiming the same access key is invisible in review and obvious to somebody driving the panel from the keyboard: the second one silently stops being reachable by its letter. It has happened, and the rule that came o...
tgeczy/panthera-speech
panthera/tests/leopard/test_settings_panel.py
.py
d18bf06e3f7d7b16
8.09
14
"""Toggling an engine setting must survive being done while speech is running. The settings reach the engine through its environment, which is read once at startup, so changing one has to replace the host process. How that replacement is done is the whole question. """ import time from test_leopard_driver import _wa...
tgeczy/panthera-speech
panthera/tests/leopard/test_toggle.py
.py
4a4a003941154338
8.09
14
# -*- coding: utf-8 -*- """A song is one utterance, however many chunks say-all hands it over in. The engine gives the last note of every utterance a sentence-final pitch fall, and 1.1.1's input-mode carry made each say-all chunk its own utterance -- so every verse ended on a droop that is in no score (panthera-speech...
tgeczy/panthera-speech
panthera/tests/leopard/test_tune_join.py
.py
0ec371eccf11ee12
8.09
14
# -*- coding: utf-8 -*- """Lion's engine fixtures. Scoped to this directory on purpose. Tiger's and Leopard's conftests call their fixtures exactly the same things, and pytest keeps them apart because a conftest only reaches its own folder and below. The NVDA fakes all three generations share are one level up. """ i...
tgeczy/panthera-speech
panthera/tests/lion/conftest.py
.py
f188ec1fe7e4ca1d
8.09
14
# -*- coding: utf-8 -*- """Silence must not cost you the next thing you ask for. **This is the bug that shipped in 0.95.0**, and three people described it from three directions before anything in this repository could see it: Jerry: "if I don't do anything on the computer for maybe a minute or two, then I hav...
tgeczy/panthera-speech
panthera/tests/lion/test_lion_idle.py
.py
abc75789439db6cb
7.09
14
# -*- coding: utf-8 -*- """10.7's input modes work now, and the stripping machinery stays. For two shipped releases this file guarded the opposite: `[[inpt TUNE]]` annotations were ignored (the melody died in two CF constructors the host had left as thunks) and a malformed phoneme after `[[inpt PHON]]` killed the host...
tgeczy/panthera-speech
panthera/tests/lion/test_lion_input_modes.py
.py
88c808a947149d02
8.09
14
# -*- coding: utf-8 -*- """A paragraph must not stop at 511 characters. Reported by Amir, who found the boundary from the outside and gave the exact paragraph to reproduce it with: "I've found an Alex-related text omission bug. It seems that if a paragraph contains more than 500 characters, Alex Premium 2.0 s...
tgeczy/panthera-speech
panthera/tests/lion/test_lion_long_text.py
.py
0bdbb617b9de340d
8.09
14
# -*- coding: utf-8 -*- """The engine has to render much faster than it speaks. **This file exists because nothing else in the suite could tell that Lion ran at nine tenths of real time.** Every voice spoke, every render was clean, the audio was byte-comparable with Leopard's -- and a three second announcement took t...
tgeczy/panthera-speech
panthera/tests/lion/test_lion_speed.py
.py
99de3b66a4b7a901
8.09
14
"""Tool Approval Policy resolver and store (CTR-0099, PRP-0067, UDR-0043). Three responsibilities: 1. ``resolve_require_set()`` -- maps ``TOOL_APPROVAL_MODE`` and ``TOOL_APPROVAL_REQUIRE_LIST`` to a ``frozenset[str]`` of tool names that should receive ``@tool(approval_mode="always_require")`` at registration...
motojinc25/chatwalaau
backend/src/app/agent/approval.py
.py
a890155fbb2e0579
7.42
6
"""Fact-level tracing for the tool-approval resume path (CTR-0009 / CTR-0100). Three releases in a row have fixed a provider 400 on this path from a hypothesis that a reproduction confirmed and production then refuted. This module exists so the NEXT report carries the facts instead: what the accumulator observed in th...
motojinc25/chatwalaau
backend/src/app/agent/approval_debug.py
.py
e860ffadba2768bd
7.42
6
"""Structured capability guidance for Prompt Assembly slot #3 (PRP-0120, CTR-0104 v4). Slot #1 is the Global Agent Identity, slot #2a/#2b are the User Profile / Agent Memory Blocks (each delimited by a `<user-profile>` / `<agent-memory>` tag), and slot #3 is the capability / tool guidance. Before PRP-0120 slot #3 was ...
motojinc25/chatwalaau
backend/src/app/agent/capability_guidance.py
.py
f14fb992551db893
7.42
6
"""Declarative agent discovery + inventory + CORE (CTR-0142, PRP-0094, UDR-0072). The bundled CORE agent reproduces current ChatWalaʻau behavior (all-``None`` spec) and is always present and active by default (UDR-0072 D7/D13). Custom agents are discovered from ``DECLARATIVE_AGENTS_DIR`` (nested folders allowed) throu...
motojinc25/chatwalaau
backend/src/app/agent/declarative/loader.py
.py
322e564c2274681f
7.42
6
"""Parse + validate + map a declarative agent YAML (CTR-0142, PRP-0094, UDR-0072). The YAML is a SPECIFICATION; ChatWalaʻau OWNS construction (UDR-0072 D1/D2). This module: 1. Parses the YAML document (best-effort schema validation via the MAF ``AgentFactory`` when ``agent-framework-declarative`` is installed, els...
motojinc25/chatwalaau
backend/src/app/agent/declarative/mapping.py
.py
13dc68a978d5677c
7.42
6
"""Declarative agent mapped spec + error types (CTR-0142, PRP-0094, UDR-0072). A ``DeclarativeAgentSpec`` is the ChatWalaʻau-side projection of a declarative agent YAML: only the fields ChatWalaʻau can MAP onto its own construction path (CTR-0070 / CTR-0102 / the prompt-assembly slots). Everything ChatWalaʻau owns -- ...
motojinc25/chatwalaau
backend/src/app/agent/declarative/spec.py
.py
ff49972a843c3d46
7.92
6
"""In-memory active declarative-agent store (CTR-0142, PRP-0094, UDR-0072 D7). Holds the single active agent id, process-global and NEVER persisted: on restart the active agent re-initializes to CORE, so the default state reproduces pre-PRP-0094 behavior byte-for-byte (UDR-0072 D7/D13). A small ``threading.Lock`` guar...
motojinc25/chatwalaau
backend/src/app/agent/declarative/store.py
.py
c4edde347b726121
7.42
6
"""Stable tool identifiers for the per-agent tool allow-list (PRP-0117, UDR-0100). A declarative agent may carry a ``tool_allowlist`` -- a list of stable identifiers that SUBSETS the globally available tool surface (UDR-0100 D1). This module is the ONE definition of that identifier scheme, shared by: * the mapping ...
motojinc25/chatwalaau
backend/src/app/agent/declarative/tool_ids.py
.py
b9ab199885ed71b2
7.42
6
"""Agent tool inventory: the selectable tool surface for authoring (CTR-0178). Composes the tool surface a declarative agent may put in its ``tools:`` allow-list into ONE read model for the authoring GUI (PRP-0117, FEAT-0061, UDR-0100 D3): * built-in FUNCTION tools -- enumerated from a static registry that mirrors ...
motojinc25/chatwalaau
backend/src/app/agent/declarative/tool_inventory.py
.py
09864de3d52ad328
7.42
6
"""Visibility into the harness conversation's own history (PRP-0148 Section 4). RES-0003 F6: the request trace reports the message list AFTER every middleware has run, fused into one ``maf_messages=`` number. Nothing reported what the history provider HELD, what it LOADED, or which session it read from -- which is why...
motojinc25/chatwalaau
backend/src/app/agent/harness/history_debug.py
.py
3cb1823316857b10
7.42
6
"""Harness agent discovery + inventory (CTR-0192, PRP-0135, UDR-0119). Discovers ``kind: Harness`` YAML from ``DECLARATIVE_AGENTS_DIR`` -- the same tree as Prompt agents and Workflows, dispatched by top-level ``kind`` through the single ``read_top_kind()`` seam (UDR-0119 D1) -- via the loader's realpath jail (the CTR-...
motojinc25/chatwalaau
backend/src/app/agent/harness/loader.py
.py
75776ee61f7766f4
7.42
6
"""Parse + validate + map a ``kind: Harness`` YAML (CTR-0192, PRP-0135, UDR-0119). The schema is ChatWalaʻau-OWNED (UDR-0119 D1): no MAF declarative factory is involved. Structural problems and unknown fields are recorded as warnings; ANY warning blocks the agent from becoming a selectable run-target (UDR-0119 D8). Cr...
motojinc25/chatwalaau
backend/src/app/agent/harness/mapping.py
.py
aa5262bc21e044f8
7.42
6
"""Harness Agent Management + Authoring API (CTR-0194 / CTR-0195, PRP-0135). CTR-0194 (management + run-target): inventory / detail / dry-run validate; the run-target is selected per-conversation via AG-UI ``state.harness_id`` (CTR-0009), NOT a global activation -- there is no rebuild and no AgentRegistry involvement ...
motojinc25/chatwalaau
backend/src/app/agent/harness/router.py
.py
70d6e3d72c563822
7.42
6
"""Per-conversation harness session cache (CTR-0009 v-next, PRP-0135, UDR-0119 D3/D4). A harness agent is a per-conversation RUN-TARGET. Its live state -- the MAF InMemoryHistoryProvider history, todo list, mode, and loop state -- lives in ONE (agent, AgentSession) pair per (harness_id, thread_id), cached here so it s...
motojinc25/chatwalaau
backend/src/app/agent/harness/runtime.py
.py
db51257110bcba48
7.42
6
"""Workspace shell executor for harness agents (CTR-0193, PRP-0135, UDR-0119 D7). WHY THIS EXISTS -- measured, not inferred: MAF's ``LocalShellTool`` executes commands through ``asyncio`` subprocess transports. Under uvicorn the running loop is a ``SelectorEventLoop`` (``uvicorn/loops/asyncio.py`` returns ``asyncio.S...
motojinc25/chatwalaau
backend/src/app/agent/harness/shell.py
.py
46f710c2041df183
7.42
6
"""Harness agent mapped spec + error types (CTR-0192, PRP-0135, UDR-0119). A ``HarnessAgentSpec`` is the ChatWalaʻau-side projection of a ``kind: Harness`` YAML -- a schema ChatWalaʻau OWNS end to end (UDR-0119 D1; MAF declarative has no harness kind). Unlike the Prompt lane (UDR-0072 D1/D2), the runtime agent IS MAF-...
motojinc25/chatwalaau
backend/src/app/agent/harness/spec.py
.py
fd9073945fc22994
7.92
6
"""Global Agent Identity and Prompt Assembly (PRP-0073, CTR-0104, UDR-0049). The agent identity -- the instance-wide persona, tone, base posture, and behaviors-to-avoid shared across every project and conversation -- is the FIRST block of every agent's system prompt ("Prompt Assembly slot #1", UDR-0049 D4). It is loa...
motojinc25/chatwalaau
backend/src/app/agent/identity.py
.py
19613fbaa9093d28
7.42
6
"""LLM prompt dump for debugging / observability (PROMPT_DUMP_*, CTR-0009 / CTR-0006). When ``PROMPT_DUMP_ENABLED`` is set (default false), each AG-UI run writes the fully assembled system prompt (Identity slot #1 + User Profile slot #2a + Agent Memory slot #2b + capability guidance) together with the run's input mess...
motojinc25/chatwalaau
backend/src/app/agent/prompt_dump.py
.py
88249d903c6d2b13
7.42
6
"""Tool Approval REST router (CTR-0099 part 2, PRP-0067). Single mutating endpoint ``POST /api/tool-approval`` that the SPA uses to release a parked tool call from the AG-UI streaming endpoint. Lifecycle: 1. AG-UI ``_stream_with_reasoning`` sees a ``function_approval_request`` content, registers an ``ApprovalReco...
motojinc25/chatwalaau
backend/src/app/agent/router.py
.py
e0adb7cc564144a8
7.42
6
"""Temporary Chat session policy (CTR-0106, PRP-0076, UDR-0052). An ephemeral, "incognito-style" conversation that: - is excluded from the user-facing history (sidebar list / search) -- it lives in a SEPARATE ``.temporary/`` quarantine directory, never in ``.sessions/``; - runs de-personalized -- the effective syst...
motojinc25/chatwalaau
backend/src/app/agent/temporary.py
.py
8eb2f5650eb09039
7.42
6
"""API Authentication Dependency (CTR-0083, PRP-0045, PRP-0051, PRP-0057). Shared FastAPI dependencies for authentication across REST endpoints, the AG-UI SSE endpoint, and the OpenAI Responses API. Two dependencies are exposed: - ``verify_api_key`` -- write endpoints consumed by the SPA, the CLI client, and AG-UI...
motojinc25/chatwalaau
backend/src/app/auth/__init__.py
.py
bde7709bc577b5d8
7.42
6
"""Password hashing for the web SPA authentication lane (CTR-0093, PRP-0057). Uses Python stdlib ``hashlib.scrypt`` -- no new package dependency. Hash format (single-line, deliberately small and grep-able): scrypt$N=<n>,r=<r>,p=<p>$<base64-salt>$<base64-hash> Constraints validated by ``parse_password_hash``: -...
motojinc25/chatwalaau
backend/src/app/auth/password.py
.py
01a372b6b0f3f958
7.42
6
"""Opaque session token store with digest-only persistence (CTR-0095 v2). Single in-memory dict guarded by an asyncio lock, projected to disk as a JSON file that holds ``sha256(token)`` digests -- never the raw token (PRP-0110 Part 2, UDR-0089 D2). A disclosed store file therefore yields no forgeable cookie: the diges...
motojinc25/chatwalaau
backend/src/app/auth/session_store.py
.py
311a87179b8346d0
7.42
6
"""Web Authentication REST API (CTR-0094, PRP-0057). Provides four endpoints for the SPA web auth lane: - ``POST /api/auth/login`` (unauth) -- exchange ID/PW for an opaque session cookie - ``POST /api/auth/logout`` (auth via CTR-0083) -- revoke the current session - ``GET /api/auth/me`` (auth via CTR-0083) -- i...
motojinc25/chatwalaau
backend/src/app/auth/web_auth.py
.py
bebc6b577d11b0cb
7.42
6
"""Shared Azure OpenAI credential resolution (PRP-0058, PRP-0059, UDR-0034). Centralizes the credential decision for every Azure OpenAI client in the runtime so that all consumers (MAF AgentRegistry, MAF DevUI agent, Whisper STT, image generation, RAG ingest embedder, RAG query embedder) agree on which lane is active....
motojinc25/chatwalaau
backend/src/app/azure_credential.py
.py
7c091c99fa556b64
7.42
6
"""In-process background task runner (CTR-0108, PRP-0077, UDR-0053). A thin, fire-and-forget asyncio runner that executes registered background tasks ON THE MAIN FastAPI event loop, AFTER a chat turn, with total error isolation. It exists for short-lived, post-conversation, LLM-driven side work that must NOT block or ...
motojinc25/chatwalaau
backend/src/app/background/__init__.py
.py
7c094c301f191ad4
7.42
6
"""Auto Session Title background task (CTR-0109, PRP-0077, UDR-0053). Summarizes a chat's first user message + first assistant reply into a concise title, replacing the leading-characters truncation. Runs only as a CTR-0108 background task (post-turn, fire-and-forget, error-isolated), so it never blocks or fails the c...
motojinc25/chatwalaau
backend/src/app/background/session_title.py
.py
eecf025bea33207b
7.42
6
"""Opt-in import gate for tool bundles that are absent from the default surface. FastMCP registers decorated tools eagerly while their modules are imported. To keep optional tool schemas out of the default MCP context, module selection must therefore happen before import rather than by removing tools afterwards. """ ...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/optional_tools.py
.py
3a6f70c350060cdd
7.56
12
import os import re from dataclasses import asdict, dataclass from functools import wraps from server.cli.runner import ( BROWSER_DEFAULT_TIMEOUT, CliAuthError, CliBrowserAuthError, CliBrowserCaptchaError, CliBrowserError, CliBrowserProfileError, CliError, CliNotFoundError, CliRegis...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/__init__.py
.py
339bba298d128e9b
7.56
12
"""MCP tools for ad extensions management.""" from server.main import mcp from server.tools import get_runner, handle_cli_errors from server.tools.helpers import ( CliOption, append_cli_options, append_pagination, check_batch_limit, run_single_id_batch, tool_error_dict, ) ADEXTENSION_GET_OPTIO...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/adextensions.py
.py
1dc41c69801a1540
7.56
12
"""MCP tools for ad video management.""" from server.main import mcp from server.tools import ToolError, get_runner, handle_cli_errors from server.tools.helpers import ( append_pagination, normalize_optional_str, require_non_empty_csv, tool_error_dict, ) @mcp.tool( name="advideos_get", descri...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/advideos.py
.py
f3ea72874df66667
7.56
12
"""MCP tools for audience target management.""" from server.main import mcp from server.tools import ToolError, get_runner, handle_cli_errors from server.tools.helpers import ( CliOption, append_cli_options, append_pagination, run_set_bids, run_single_id_batch, tool_error_dict, ) AUDIENCE_GET_...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/audience.py
.py
3befe41724b72ada
7.56
12
"""MCP tools and prompts for direct auth profiles.""" import json import time from mcp.server.fastmcp import Context from pydantic import BaseModel, Field from server.cli.runner import ( CliError, CliNotFoundError, CliTimeoutError, DirectCliRunner, _find_direct, _strip_ansi, ) from server.mai...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/auth_tools.py
.py
858df939eb21b671
7.56
12
"""MCP tools for checking changes in Yandex.Direct.""" import re from server.main import mcp from server.tools import ToolError, get_runner, handle_cli_errors from server.tools.helpers import check_batch_limit, parse_ids, tool_error_dict ALLOWED_FIELD_NAMES = {"CampaignIds", "AdGroupIds", "AdIds", "CampaignsStat"} ...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/changes.py
.py
316388bd73cd48e8
7.56
12
"""MCP tools for creatives management.""" from server.main import mcp from server.tools import get_runner, handle_cli_errors from server.tools.helpers import append_pagination @mcp.tool( name="creatives_get", description="List creatives (rich media/video creatives built in Constructor), optionally filtered b...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/creatives.py
.py
de100cb4fe5fcd26
7.56
12
"""MCP tools for Yandex.Direct dictionaries.""" from server.main import mcp from server.tools import get_runner, handle_cli_errors # Names accepted by Dictionaries.get in Direct API v5. The CLI's own # `dictionaries list-names` only knows the first ten, but `dictionaries get` # (and the API) accept the audience / sch...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/dictionaries.py
.py
295a6f1cb4bdd09d
7.56
12
"""MCP tools for dynamic ad (webpage) management.""" from server.main import mcp from server.tools import get_runner, handle_cli_errors from server.tools.helpers import ( append_id_filters, append_pagination, run_set_bids, run_single_id_batch, ) @mcp.tool( name="dynamicads_get", description="...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/dynamic_ads.py
.py
000cc28f22a6ef4b
7.56
12
"""MCP tools for dynamic feed ad target management.""" from server.main import mcp from server.tools import get_runner, handle_cli_errors from server.tools.helpers import ( append_id_filters, append_pagination, run_set_bids, run_single_id_batch, tool_error_dict, validate_yes_no, ) @mcp.tool( ...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/dynamic_feed_ad_targets.py
.py
0f65b16991ff1c12
7.56
12
"""MCP tools for feed management.""" from server.main import mcp from server.tools import ToolError, get_runner, handle_cli_errors from server.tools.helpers import ( CliOption, append_cli_options, append_pagination, require_update_fields, run_single_id_batch, tool_error_dict, ) BUSINESS_TYPES ...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/feeds.py
.py
c142e0c4cc624874
7.56
12
"""Shared helpers for MCP tool modules.""" import json from collections.abc import Iterable, Mapping, Sequence from dataclasses import dataclass from server.cli.runner import ( CliAuthError, CliError, CliNotFoundError, CliRegistrationError, CliTimeoutError, ) from server.tools import ToolError, _h...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/helpers.py
.py
5f2b9eaae7048541
7.56
12
"""MCP tools for ad images management.""" from server.main import mcp from server.tools import ToolError, get_runner, handle_cli_errors from server.tools.helpers import ( append_pagination, normalize_optional_str, tool_error_dict, validate_yes_no, ) # adimages_add uploads a base64-encoded JPEG/PNG (re...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/images.py
.py
4ab9681e39acb1cd
7.56
12
"""Read-only MCP tools for browser-backed Campaign Wizard campaigns.""" import os from typing import cast from server.contract import PUBLIC_CONTRACT from server.main import mcp from server.tools import ToolError, get_browser_runner, handle_cli_errors from server.tools.browser_helpers import browser_session_args from...
axisrow/yandex-direct-mcp-plugin
plugins/yandex-direct/server/tools/masters.py
.py
fdfe24f6352a0ae2
7.56
12