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 |
|---|---|---|---|---|---|---|
"""Local cache for detail-page HTML.
Layout: detail_cache/<source>/<source_id>.html.
TTL: 24h by default. Re-runs the same day skip the network entirely.
"""
import time
from pathlib import Path
CACHE_DIR = Path(__file__).parent.parent.parent / "detail_cache"
DEFAULT_TTL_S = 24 * 60 * 60 # 24h
def _path(source: st... | luke-song/rent-kit | src/rentkit/cache.py | .py | 4a3c1059eb400360 | 7 | 0 |
"""GCS-canonical Rentkit DB.
Private deployments can mirror the SQLite DB to a GCS object. Every write
verb pulls a fresh copy, mutates locally, and pushes back with an
if-generation-match precondition. If the precondition fails, the caller can
pull once more and retry the verb.
Read-only verbs can use `with_db(read_... | luke-song/rent-kit | src/rentkit/cloud_sync.py | .py | 66ca77d1e994ca4b | 7 | 0 |
"""Craigslist scraper.
No bot wall to speak of. Uses the configured sub-area searches, filtered by
the bedroom range and dog flag from `search.toml`. Drops the neighborhood-id
filter on purpose: codes drift, and the result set is small enough to filter
on the title/hood text client-side.
"""
import re
from playwright... | luke-song/rent-kit | src/rentkit/craigslist.py | .py | d9f695a4bbc2d038 | 7 | 0 |
"""Cross-source dedup.
Two listings are the same property if:
- both have lat/lng and they're within ~80m, AND
- bed counts match (or one is None), AND
- prices are within 20%.
Fallback when lat/lng is missing on one side: normalized street-address match.
When merging a cluster, fields are coalesced in source priori... | luke-song/rent-kit | src/rentkit/dedup.py | .py | 8e9cd185fbb544b3 | 7 | 0 |
"""Local mirror of listing photos.
Each listing's photos get downloaded once to tmp/photos/<source>/<source_id>/N.jpg
and the listing.photos URLs are rewritten to relative paths the Firebase host
serves directly. This protects against Zillow CDN expiry and source-page delisting.
"""
import asyncio
import os
from pathl... | luke-song/rent-kit | src/rentkit/photos.py | .py | cfca8fe3e3906fde | 7 | 0 |
"""Rank listings by fit.
Heuristic baseline. Higher score = better.
Inputs are weighted in line with stated priorities, in priority order:
- Dogs OK (large or any-size) — gate; no-dogs heavily penalized
- Walk-to-trail — primary
- Walk-to-beach — secondary
- 3 bedrooms preferred, ≥ 1.5 baths preferred
- In-... | luke-song/rent-kit | src/rentkit/rank.py | .py | 1eecb6e760964ef0 | 7 | 0 |
"""Redfin scraper.
Redfin sits behind PerimeterX (same wall as Zillow), so we drive the shared
persistent Playwright profile (`.chrome-profile`) — a captcha cleared once via
`rentkit solve`-style flow sticks across runs. The rental search server-renders
the listing cards into the page (the data is NOT in an XHR), so w... | luke-song/rent-kit | src/rentkit/redfin.py | .py | ddcd79566fc9c12c | 7 | 0 |
"""The search config: where you're looking, and what you want.
Everything market-specific used to live as literals spread across the
scrapers, the geocoder, the anchor lists, the scorer, and the LLM prompts.
This module is the one place that knows those answers, read from a
`search.toml` file so a new city is a config... | luke-song/rent-kit | src/rentkit/search_config.py | .py | 4df4c7e652f48a79 | 7 | 0 |
"""What a source run actually covered.
A blocked search and an empty search both used to return `[]`. `search` reads
that list twice: once to upsert what it found, and once — through
`succeeded_sources` — to decide which listings are gone from the market and
should be marked inactive.
That second read is where the di... | luke-song/rent-kit | src/rentkit/sources.py | .py | fe83a1c71da6d83b | 7 | 0 |
"""SQLite-backed storage for listings.
One row per (source, source_id). Each search upserts; `last_seen` tracks
freshness, `active` marks whether the most recent run saw the listing.
"""
import json
import os
import sqlite3
from contextlib import contextmanager
from datetime import datetime
from pathlib import Path
f... | luke-song/rent-kit | src/rentkit/storage.py | .py | 92de6a02239b6c29 | 7 | 0 |
"""Walking-time estimates from a listing to named anchors.
The anchors come from the search config, in four groups:
- BEACHES / TRAILS / BAKERIES: the "somewhere worth walking to" set. Each
group is collapsed to its nearest member, so adding an anchor to the
config widens the search rather than adding anothe... | luke-song/rent-kit | src/rentkit/walk.py | .py | 838ecabb57ccccf1 | 7 | 0 |
"""Zillow scraper.
Search results: parse `__NEXT_DATA__` JSON from the search page (clean,
structured, doesn't need a real browser render).
Detail enrichment: Zillow embeds the facts as a flat list of "Key: Value"
items inside `<ul class*=Fact> <li>`. We pull the list raw and key off it
instead of regex-grepping the ... | luke-song/rent-kit | src/rentkit/zillow.py | .py | 27c109e758f772ca | 7 | 0 |
"""Zumper scraper.
Zumper hydrates the search results into `window.__PRELOADED_STATE__`
under `currentSearch.listables.listables` — same shape-trick as Zillow's
`__NEXT_DATA__` but stashed on the window instead of a script tag. We
pull the array directly via `page.evaluate` and skip DOM parsing.
The configured URL fi... | luke-song/rent-kit | src/rentkit/zumper.py | .py | 50cadf015ecc18e0 | 7 | 0 |
"""A cloud-synced verb without a bucket should explain itself, not traceback.
Cloud sync is optional plumbing, but the verbs that use it default to it. On a
single-machine setup — which is every setup until someone deliberately sets up
GCS — the first `rentkit why` used to end in a RuntimeError raised several
frames i... | luke-song/rent-kit | tests/test_cloud_not_configured.py | .py | 7e97dc60bd4208ad | 7.5 | 0 |
"""A source that could not look has not established that anything is gone.
`storage.upsert_run` marks a listing inactive when a source reported in and the
listing was not among what it reported. That inference is only sound if the
source actually read everywhere it searches. Blocking happens per area; the
guard was pe... | luke-song/rent-kit | tests/test_partial_coverage.py | .py | 0720bb369eba3ad8 | 7.5 | 0 |
"""A score of zero has two causes, and the integer cannot tell them apart.
`score_detail` mirrors `score` term for term. The mirroring is the risk in this
change, so the first test here is the one that holds the two together across
the whole fixture.
"""
import shutil
import rentkit
from rentkit import storage, walk... | luke-song/rent-kit | tests/test_score_detail.py | .py | a9a99ff053f52c20 | 7.5 | 0 |
"""The config is now load-bearing, so it gets tested like code.
Every market literal used to live next to the code that consumed it, where a
typo broke one scraper loudly. Now one file feeds the scrapers, the geocoder,
the anchors, the scorer and the prompts, so a wrong value is wrong everywhere
at once and a missing ... | luke-song/rent-kit | tests/test_search_config.py | .py | 04de4a1f64fbc254 | 7.5 | 0 |
import pytest
from rentkit import craigslist, dedup
from rentkit.locations import PRIMARY_TERMS, SECONDARY_TERMS
def test_source_priority_keeps_zillow_as_primary_source():
assert dedup.SOURCE_PRIORITY["zillow"] == 0
assert dedup.SOURCE_PRIORITY["zillow"] < dedup.SOURCE_PRIORITY["craigslist"]
def _respaced(... | luke-song/rent-kit | tests/test_source_config.py | .py | 19cb13a99b9c742a | 7.5 | 0 |
"""A weight table is a claim about the scorer, so it gets checked like one.
`TERM_WEIGHTS` is what `rentkit why` prints when it says what the policy weighs
most. Hand-written tables drift from the code they describe and then report
something nobody verified — which is the failure this whole branch is about.
These test... | luke-song/rent-kit | tests/test_term_weights.py | .py | ada44ea3921b1550 | 7.5 | 0 |
"""The output is the feature, so the output gets tested.
`score_detail` returning correct lists is not the change — a person reading a
screen and coming away with the right belief is. These tests pin the sentences
that carry that, because a refactor that silently drops "never evaluated" turns
the tool back into the th... | luke-song/rent-kit | tests/test_why_output.py | .py | 2c94f05e96042609 | 7.5 | 0 |
"""Top 10 candidates — domain-colored bar chart with track-colored gene names."""
from __future__ import annotations
import sys; sys.path.insert(0, str(__import__("pathlib").Path(__file__).resolve().parent))
from style import *
import matplotlib.pyplot as plt, numpy as np, pandas as pd
from matplotlib.patches import Pa... | dhzso/neuraltf | projects/NeuralTF/scripts/figures/05_top10_candidate_atlas.py | .py | c6269079317949aa | 7.15 | 1 |
"""Composite bonus waterfall — shows base score + each bonus for top-10."""
from __future__ import annotations
import sys; sys.path.insert(0, str(__import__("pathlib").Path(__file__).resolve().parent))
from style import *
import matplotlib.pyplot as plt, numpy as np, pandas as pd
GO_NEURAL_KW = {"neurogenesis","nervou... | dhzso/neuraltf | projects/NeuralTF/scripts/figures/18_composite_bonus_waterfall.py | .py | d0d9408f2bedff79 | 7.15 | 1 |
"""Validate NeuralTF top-10 candidates against Perez 2025 ANANSE predictions.
Cross-references our prioritized TFs against the ANANSE-predicted
TF-target regulatory network from Perez et al. 2025 (MOESM22).
"""
from __future__ import annotations
import sys
from pathlib import Path
import pandas as pd
ROOT = Path(__... | dhzso/neuraltf | projects/NeuralTF/scripts/validate_with_perez.py | .py | dba8ba61a20c9a96 | 7.15 | 1 |
#!/usr/bin/env python
"""Permutation baseline for NeuralTF pipeline.
Shuffles cluster labels in scRNA-seq atlases to generate null distribution
of integrated scores. Computes empirical p-values for real candidates.
Usage:
python scripts/permutation_baseline.py --n-perm 10 --subsample 2000
"""
import argparse
imp... | dhzso/neuraltf | scripts/permutation_baseline.py | .py | e8eb53b36ed16d0c | 7.15 | 1 |
"""Generate concise, evidence-backed portfolio findings from the loaded real dataset."""
from __future__ import annotations
import argparse
import sqlite3
from pathlib import Path
ROOT = Path(__file__).resolve().parents[1]
def rows(connection, query):
return connection.execute(query).fetchall()
def main() -> int... | dytcoke23/ecommerce-revenue-customer-experience-analytics | src/generate_reports.py | .py | 7b301508bbddf565 | 7 | 0 |
# -*- coding: utf-8 -*-
"""
文档速读工具 read_files.py
按扩展名自动分派最优提取器，批量输出结构化文本。
用法:
python read_files.py <path> [path...]
path 可以是文件或文件夹（文件夹会递归处理）
python read_files.py --markitdown <path> # PDF/docx/pptx/图片 走 MarkItDown 统一转换
python read_files.py --mineru <path> # PDF 走 MinerU 深度解析（Windows 命令行不稳）
工具映射:
.... | GZLns/Document-reader- | read_files.py | .py | 1e4a871831922754 | 7 | 0 |
# -*- coding: utf-8 -*-
"""
xlsx/xls/csv 速查工具 xlsx_query.py（2026-08-25 升级版）
读取层：python-calamine（Rust 引擎，支持 .xlsx/.xls/.ods，速度远超 openpyxl）
查询层：原生细粒度查询 + duckdb SQL 模式（直查 xlsx/csv，无需导入）
用法:
python xlsx_query.py <file> --list # 列出所有 sheet 和尺寸
python xlsx_query.py <file> --sheet Sheet0 --head 10 ... | GZLns/Document-reader- | xlsx_query.py | .py | ec1eb78493d876a1 | 7 | 0 |
# -*- coding: utf-8 -*-
"""
仓库级 PLC 规范审查（CI 门禁）。
检查项（对应 AGENTS.md R1~R4 + 制表/编码规则）：
1. 编码：CSV 必须 GBK，.md 必须 UTF8（无 BOM）
2. R1 标识符仅英文：扫描 .st 代码文件中的中文字符
3. 魔法数字检查：代码中裸数字字面量告警（白名单除外）
4. CSV 列数一致性：每行字段数 = 表头字段数（拦截"含逗号字段未加引号"导致的列错位）
5. 通讯字表专项：5 列结构、Mobus=40001+offset、写1触发段标记
6. 通讯字表 ↔ host 变量表一致性：host_Rcv_* 的 D ... | TReaur1/ELE | scripts/ci_check.py | .py | 531cd3813f1f626d | 7 | 0 |
# -*- coding: utf-8 -*-
"""一键生成 + 审查.
用法: python scripts/generate.py <project>
流程: 加载规格单 -> 生成表格 -> 生成ST -> 审查ST
若审查发现未声明符号/R1违规, 返回非0退出码.
"""
import os, subprocess, sys
BASE = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
def run(script, *args):
r = subprocess.run([sys.executable, os.path.join(BAS... | TReaur1/ELE | 设备模型库/scripts/generate.py | .py | b998c2b93c1cd085 | 7 | 0 |
# -*- coding: utf-8 -*-
"""生成代码审查 (review_st): 校验生成 ST 是否符合 AGENTS.md 规则.
检查项:
R1 标识符仅英文(注释外的中文)
未声明符号(变量/实例/常量/类型 须在表中声明)
FB 结构平衡(VAR_INPUT/OUTPUT/IN_OUT/END_VAR)
CASE 选择器为整型
魔法数字(数值字面量)
常见缺陷清单(见 AGENTS.md 六)
"""
import os, re, sqlite3, sys
BASE = os.path.dirname(os.path.dirname(os.path.abspath(__file__))... | TReaur1/ELE | 设备模型库/scripts/review_st.py | .py | fa634efa1fee343e | 7 | 0 |
"""Windows-headless / detached process-spawn helpers (shared, vendored per plugin).
A copilot-extensions runtime frequently runs **windowless**: under a hidden
Windows Scheduled Task launched via ``conhost --headless``, inside the
agent-bridge daemon, or from a session-start hook whose parent has no console of
its own... | ThomasMichon/copilot-extensions | libs/agent-procutil/src/agent_procutil/__init__.py | .py | 8a61fc73adc72264 | 7.24 | 2 |
"""Core migration primitives: ``migrate_doc`` (in-memory) and ``migrate_file``.
Two call sites, one migrator registry -- the *migrate-by-rewrite* model:
* ``migrate_doc`` applies the ordered ``vN->vN+1`` migrators to a parsed config
document **in memory** and returns ``(new_doc, changed)``. This is the
**loader's... | ThomasMichon/copilot-extensions | libs/config-migrate/src/config_migrate/core.py | .py | 77d2469a245ee27f | 7.24 | 2 |
"""Schema registry: schema_id -> current version + ordered vN->vN+1 migrators.
A ``SchemaRegistry`` is the single source of shape-change truth for a set of
managed config files. Each managed schema declares:
* a stable ``schema_id`` (e.g. ``"agent-worktrees/config"``),
* a monotonic integer ``current_version`` (match... | ThomasMichon/copilot-extensions | libs/config-migrate/src/config_migrate/registry.py | .py | e542436a67d0ca48 | 7.24 | 2 |
"""Runner: migrate a set of managed config files against a registry.
The *discovery* of which files are managed (by convention from an install root)
is deliberately left to the consuming plugin -- only the plugin knows its own
config layout. The runner takes an explicit list of ``ManagedFile`` entries and
applies ``mi... | ThomasMichon/copilot-extensions | libs/config-migrate/src/config_migrate/runner.py | .py | 5d5ab7c743567c5b | 7.24 | 2 |
"""Provider source-injection API for the credential relay.
agent-bridge runs the relay in its daemon and discovers provider plugins that
inject credential sources for their targets. Each provider exposes::
# <provider_pkg>/relay_provider.py
def register_relay(builder: RelayBuilder) -> None:
builder.ad... | ThomasMichon/copilot-extensions | libs/credential-relay/src/credential_relay/registry.py | .py | 89046008e218fc0f | 7.24 | 2 |
"""Pluggable credential resolution sources.
Each source handles a subset of credential requests (by action and/or
host pattern). The relay server routes requests to the first source
whose ``supports()`` returns True.
"""
from __future__ import annotations
from typing import Protocol, runtime_checkable
@runtime_che... | ThomasMichon/copilot-extensions | libs/credential-relay/src/credential_relay/sources/__init__.py | .py | a63b123e4ea4888b | 7.24 | 2 |
"""Azure CLI token source.
Returns Azure access tokens via ``az account get-access-token``.
Handles the ``get-azure-token`` relay action, returning the token in
git-credential-protocol key=value format for uniform framing.
This is a HIGH-TRUST credential source. It must be explicitly enabled
and configured with an ex... | ThomasMichon/copilot-extensions | libs/credential-relay/src/credential_relay/sources/az_login.py | .py | 3364aa3f90a8e74d | 7.24 | 2 |
"""GitHub CLI auth token source.
Returns ``gh auth token`` output for GitHub hosts. Handles the
``get-github-token`` relay action, returning the token in
git-credential-protocol key=value format for uniform framing.
"""
from __future__ import annotations
import asyncio
import logging
import subprocess
import sys
lo... | ThomasMichon/copilot-extensions | libs/credential-relay/src/credential_relay/sources/gh_auth.py | .py | 2b59fa78243414ce | 7.24 | 2 |
"""Git Credential Manager (GCM) proxy source.
Proxies credential requests to the local ``git credential`` command,
which typically resolves through Git Credential Manager. Includes
WSL detection (routes through PowerShell when running under WSL),
credential caching with TTL, and request coalescing for expensive
GCM ro... | ThomasMichon/copilot-extensions | libs/credential-relay/src/credential_relay/sources/git_credential.py | .py | 0940520b086d77e8 | 7.24 | 2 |
"""Injected Azure-token source (test/relay-shim affordance).
Serves the ``get-azure-token`` action from a **pre-minted bearer supplied in an
environment variable**, instead of shelling ``az account get-access-token`` like
:class:`~credential_relay.sources.az_login.AzLoginSource`.
Why this exists
---------------
The r... | ThomasMichon/copilot-extensions | libs/credential-relay/src/credential_relay/sources/injected_token.py | .py | 0ed8198b0c1d9e6b | 7.24 | 2 |
"""Tests for the injected Azure-token relay source (test/relay-shim affordance).
Covers the production-inert passthrough contract and the serve/allowlist/expiry
behavior. The build-wiring assertion (injected source placed BEFORE az-login)
lives in test_relay_shim.py, which has the relay_token isolation fixture.
"""
f... | ThomasMichon/copilot-extensions | libs/credential-relay/tests/test_injected_token.py | .py | 0958152529408f38 | 7.74 | 2 |
"""Tests for relay-port reclaim on a stale holder (#19)."""
from __future__ import annotations
import asyncio
import errno
import os
import socket
import subprocess
import sys
from credential_relay.server import (
CredentialRelayServer,
_addr_in_use,
_pid_on_port,
_reclaim_port,
)
def _free_port() ... | ThomasMichon/copilot-extensions | libs/credential-relay/tests/test_port_reclaim.py | .py | 9a3004d863526016 | 7.74 | 2 |
"""Structured scan and entry verdicts shared by drop-in registry consumers."""
from __future__ import annotations
import hashlib
import json
from collections.abc import Mapping
from dataclasses import dataclass, field
from enum import Enum
from typing import Generic, TypeVar
T = TypeVar("T")
class ScanAuthority(st... | ThomasMichon/copilot-extensions | libs/dropin-registry/src/dropin_registry/model.py | .py | d290082f3c5c5088 | 7.24 | 2 |
"""Bounded operational-warning selection."""
from __future__ import annotations
import time
from collections.abc import Iterable
from dataclasses import dataclass, field
from .model import Finding
@dataclass(frozen=True)
class WarningBatch:
"""Detailed findings selected for one operational emission."""
em... | ThomasMichon/copilot-extensions | libs/dropin-registry/src/dropin_registry/warnings.py | .py | 922eb1b43fd456cd | 7.24 | 2 |
"""Rendezvous (port-mapping) files for discoverable, collision-free local endpoints.
A *service-bearing* Copilot CLI plugin needs its clients -- its own CLI, sibling
plugins, and agents on the box -- to reach it without hardcoding a fixed loopback
TCP port. Pinning a port collides with siblings, with the ``127.0.0.1``... | ThomasMichon/copilot-extensions | libs/endpoint-rendezvous/src/endpoint_rendezvous/rendezvous.py | .py | 7ae999b9b2ec81a4 | 7.24 | 2 |
"""Load a plugin **marketplace** manifest and resolve a plugin's on-disk source
directory, across the Copilot-native and Claude conventions.
A marketplace directory carries a ``marketplace.json`` (at ``.github/plugin/``,
the root, ``.plugin/``, or ``.claude-plugin/`` -- native-first) listing plugins,
each with a ``sou... | ThomasMichon/copilot-extensions | libs/plugin-resolve/src/plugin_resolve/marketplace.py | .py | 847f6dbf1a64d6d2 | 7.24 | 2 |
#!/usr/bin/env python3
"""Hermes shell hook: convention + failure signal recorder (post_tool_call).
⚠️ IMPORTANT — ``post_tool_call`` results are OBSERVATIONAL in Hermes: the
shell-hook bridge parses ``{"context": ...}`` for any event, but the sole
emitter (``model_tools._emit_post_tool_call_hook``) **discards the ret... | zaxbysauce/zmem | hermes-plugin/hooks/zmem-hermes-convention.py | .py | 86a9e5de86d9a6c9 | 7 | 0 |
#!/usr/bin/env python3
"""Hermes shell hook: reflect nudge DELIVERY (pre_llm_call) — universal surface.
This is the DELIVERY side of the reflection loop. The convention hook
(``post_tool_call``, observational — its results are discarded by Hermes)
records pending-nudge flags in zmem's ``meta`` table; THIS hook fires o... | zaxbysauce/zmem | hermes-plugin/hooks/zmem-hermes-reflect.py | .py | 8ade08d420518ea1 | 7 | 0 |
#!/usr/bin/env python3
"""Hermes shell hook: coding-stop reflect nudge (pre_verify).
Fires only on coding turns where the agent edited files and is about to
verify/finish (Hermes gates ``pre_verify`` on ``_turn_file_mutation_paths``,
so non-coding gateway sessions never reach this hook). Emits
``{"action": "continue",... | zaxbysauce/zmem | hermes-plugin/hooks/zmem-hermes-verify.py | .py | 636dc228dd013600 | 7 | 0 |
"""SDK-native Bearer-token auth for the zmem MCP server.
Uses the ``mcp`` SDK's built-in ``TokenVerifier`` Protocol rather than custom
Starlette middleware. The SDK wires ``BearerAuthBackend`` +
``RequireAuthMiddleware`` automatically when ``FastMCP(auth=..., token_verifier=...)``
is constructed (verified in mcp/serve... | zaxbysauce/zmem | hermes-plugin/server/auth.py | .py | 165a997ed770e587 | 7 | 0 |
"""Bind-address guard for the zmem MCP server.
Security control (not just docs): refuse to bind to a wildcard address
(``0.0.0.0`` / ``::``) unless the operator has explicitly opted in via
``ZMEM_MCP_ALLOW_INSECURE_BIND=1``. A Bearer-protected store broadcasting on
the whole network is a real exposure — the token is t... | zaxbysauce/zmem | hermes-plugin/server/bind_guard.py | .py | 3539d23ea8c2f9f4 | 7 | 0 |
#!/usr/bin/env python3
"""Shared recall body for the injecting hooks (issue #58, 3.5/3.8/3.9).
Consumers (all invoke this file AS A SCRIPT — the hyphenated filename
cannot be imported):
- zmem-recall.sh (UserPromptSubmit) mode "user_prompt"
- zmem-precompact.sh (PreCompact, Claude only) mode "pr... | zaxbysauce/zmem | hooks/lib/zmem-recall-body.py | .py | a7bfa44570096145 | 7 | 0 |
"""Public-corpus adapters for the zmem eval gold format (issue #64, 9.3).
Converts published eval sets into the gold JSONL shape that
scripts/eval_runner.py and `store.py tune-weights` consume:
python scripts/eval_adapters.py --adapter longmemeval --input <path> --out <path>
python scripts/eval_adapters.py --... | zaxbysauce/zmem | scripts/eval_adapters.py | .py | 8bc4bf51c74f2359 | 7 | 0 |
"""Offline eval runner for the zmem memory store (issue #64, 9.1 + 9.5).
THE canonical eval command — CI (`.github/workflows/ci.yml`) runs exactly
this, and SKILL.md documents exactly this:
python scripts/eval_runner.py --store <path> [--gold eval/gold.jsonl] \\
[--k 5] [--fail-under X] [--json-out PATH]
... | zaxbysauce/zmem | scripts/eval_runner.py | .py | 7403d89ef75e9f86 | 7 | 0 |
#!/usr/bin/env python3
"""Ingest a closeout-remote harvest JSON array into the zmem store.
Reads a harvest JSON file (the array format documented in
skills/closeout-remote/SKILL.md's "Output format" section), validates each
row's shape and enum values, and calls `store.py add` via subprocess for
each valid row.
This ... | zaxbysauce/zmem | scripts/ingest_harvest.py | .py | 4deed83a20c9227a | 7 | 0 |
#!/usr/bin/env python3
"""Pattern-tuning harness for zmem's correction-pattern library (issue #46).
Adapted from claude-reflect (https://github.com/BayramAnnakov/claude-reflect),
MIT-licensed, `scripts/compare_detection.py` — minus the semantic/LLM side.
zmem's scripts must be stdlib-only and multi-host: this harness ... | zaxbysauce/zmem | scripts/pattern_harness.py | .py | 428b33978ee59b7c | 7 | 0 |
#!/usr/bin/env python
"""Release gate: make the repo's release contract self-enforcing.
The documented contract (README / CHANGELOG) is "released versions are marked
with a git tag (vX.Y.Z) and a GitHub Release". That step was manual, and the
manual step silently rotted: v0.8.5/v0.8.6/v0.8.8 were never tagged and
v0.9... | zaxbysauce/zmem | scripts/release_gate.py | .py | e176740d5f2ddeeb | 7 | 0 |
"""Embedding profile registry (issue #63, 8.2). Single source of truth for the
model/dim/hash facts of every shipped embedding profile.
Both ``embeddings.py`` (the loader), ``storelib/schema.py`` (the vec0 DDL dim),
``storelib/cli.py`` (profile selection + dim-mismatch refusal), and
``doctor.py`` (the embeddings_healt... | zaxbysauce/zmem | skills/memory/scripts/embed_profiles.py | .py | 0666878e7c4924bf | 7 | 0 |
#!/usr/bin/env python
"""ZMem legacy-store import — Phase 1 (box-wide unified memory, PLAN.md P1).
Copies the existing per-plugin ZCode store into the new box-neutral location
(~/.zmem by default) WITHOUT ever opening the source read-write. The legacy
store may be live (an active ZCode session writing to it) so this s... | zaxbysauce/zmem | skills/memory/scripts/import-store.py | .py | 75047ada65ad486b | 7 | 0 |
from __future__ import annotations
# PRR-032 note: the from-import lines below re-import several names from
# DIFFERENT submodules (ALLOWED_SIGNALS, STORE_PATH, stdlib modules, ...).
# Python binds the LAST occurrence, so when adding a new submodule line,
# verify any name it re-exports that an earlier line also impor... | zaxbysauce/zmem | skills/memory/scripts/storelib/__init__.py | .py | 9881fab5981720d4 | 7 | 0 |
"""Optional post-MMR cross-encoder rerank (issue #63, 8.6).
POLICY (load-bearing — do not weaken):
- Default OFF everywhere. Enablement requires ZMEM_CROSS_ENCODER to opt in.
- Rerank fires ONLY on explicit CLI `recall` runs that also mutate telemetry:
`cli_allowed` demands recall-without---no-bump. Every hook surfa... | zaxbysauce/zmem | skills/memory/scripts/storelib/cross_encoder.py | .py | be9da0f236b28d38 | 7 | 0 |
from __future__ import annotations
import argparse
import calendar
import contextlib
import hashlib
import json
import math
import os
import re
import shutil
import sqlite3
import struct
import subprocess
import sys
import time
import uuid
import glob
from datetime import datetime, timezone
from pathlib import Path
fr... | zaxbysauce/zmem | skills/memory/scripts/storelib/promote.py | .py | d6f7eb5b82c31761 | 7 | 0 |
"""Build a tiny deterministic store fixture for CLI characterization.
Task 2.1 (issue #57): the characterization suite freezes stdout hashes of
`stats`, `list --json`, `recall --json` and `export-jsonl` against a fixture
store, and must fail if the public CLI surface (subcommands / required flags)
drifts. This builder... | zaxbysauce/zmem | tests/fixtures/store_builder.py | .py | 4d4c1a687e2ce60b | 7.5 | 0 |
"""Issue #58, 3.6: ``--as-of ISO-8601`` temporal predicate.
Three rows with staggered ``valid_from``; recall at T2 returns only the
first two. Absent flag → all live rows (current behavior).
Also tests the Z-suffix normalization: ``+00:00`` input must compare
correctly against ``valid_from`` stored with a Z-suffix (I... | zaxbysauce/zmem | tests/test_as_of_recall.py | .py | 7fd01ae3f8c0ddd6 | 7.5 | 0 |
"""Executable transformation-family classifiers (human side only).
These implement PREREG-CENSUS section 4 and 5. They are used to ATTACK the
grammar (is a human mutation taxonomy privileged by the physics?). They are
never visible to any learner, never enter generation, selection, routing or
admission, and are record... | jcraig949jfi/Prometheus | agent_d2_blind/d2/classify.py | .py | 8818b99fa68a508c | 7.24 | 2 |
"""G3B REWRITE-ONEPASS basis.
Identical grammar to G3. Only the rewrite strategy differs: a single pre-order
pass over the ORIGINAL positions, each position rewritten at most once and the
result never re-scanned.
Why this variant exists (disclosed multiplicity): a smoke test — run before any
census — showed that the ... | jcraig949jfi/Prometheus | agent_d2_blind/d2/g3b.py | .py | 28a81189d82673db | 7.24 | 2 |
"""Anti-cheat battery. Must pass before any evidence is read.
Static checks scan code with comments and string literals stripped, so a
docstring mentioning a forbidden word cannot pass or fail a check by itself.
"""
import io
import json
import os
import sys
import tokenize
HERE = os.path.dirname(os.path.abspath(__fi... | jcraig949jfi/Prometheus | agent_d3_blind/anti_cheat/checks.py | .py | 3ad38463dbcf5360 | 7.24 | 2 |
"""Shared M0 harness: identical physics, identical validity API, identical meter.
The baselines see ONLY the whitelisted numeric observation dict returned by
`Ctx.evaluate`. Semantic fingerprints, targets, witnesses and family labels are
recorded on the harness side and are never returned to a baseline.
"""
import ra... | jcraig949jfi/Prometheus | agent_d3_blind/m0/harness.py | .py | 5c76f4c35ec5e336 | 7.24 | 2 |
"""Frozen probe batteries.
VALUE_PROBES fix all semantic equivalence in this experiment; every equivalence
claim is therefore probe-relative and is labelled as such. EXT_PROBES exist
only for the preregistered probe-stability check.
"""
import hashlib
import random
VALUE_PROBES = [
(),
(0,),
(5,),
(1... | jcraig949jfi/Prometheus | agent_d3_blind/probes/battery.py | .py | da398ea3e6ca3f19 | 7.24 | 2 |
"""Basis registry and canonical-order control (frozen)."""
import random
from . import s1_tpc, s2_flat, s3_trs, s4_rev
BASES = {"S1": s1_tpc, "S2": s2_flat, "S3": s3_trs, "S4": s4_rev}
ORDER_SPACES = {
"S1": (s1_tpc.NSLOT, s1_tpc.NARG),
"S2": (s2_flat.NOPS, s2_flat.NARG),
"S3": (s3_trs.NKIND, s3_trs.NPAY)... | jcraig949jfi/Prometheus | agent_d3_blind/substrates/registry.py | .py | faf09da7613b6be8 | 7.24 | 2 |
"""Anti-cheat static battery. Spec: constitution section 39.
Seam doctrine: learner-visible data is ALLOWLISTED at field level. The only
learner-visible object is learner_view(task) == {'domain', 'table'}; both hold
pure integers. Anything else (family, seed, gen_meta, witness, strata, oracle
solutions) is oracle-side... | jcraig949jfi/Prometheus | agent_d5_blind/anti_cheat/static_checks.py | .py | ab391ae45a3b7383 | 7.24 | 2 |
"""
MechKernel Adaptive Renderer（M1 阶段）
C 方案的核心：决定每步操作该用哪个 render_level。
- none: 不渲染（默认）
- iso_only: 1 张 iso 视角
- full: 4 视角
策略：
1. 拓扑变化必渲染 iso_only
2. 失败恢复必渲染 iso_only
3. 间隔 N 步渲染一次 full
4. 草图/查询/状态操作不渲染
"""
from typing import Optional, Dict
import time
from .features import FeatureType, TOPOLOGY_CHANGING_OPS, NON_... | vanyu0710/mechcad-kernel | mech_kernel/adaptive_renderer.py | .py | c97070ea1127f443 | 7 | 0 |
"""
MechKernel AI Orchestrator (v1 单 orchestrator)
P2-10 (v8 DeepSeek): 加结构化日志
"""
import logging
_logger = logging.getLogger("mech_kernel.orchestrator")
if not _logger.handlers:
_h = logging.StreamHandler()
_h.setFormatter(logging.Formatter(
"[%(asctime)s] %(levelname)s orchestrator: %(message)s",
... | vanyu0710/mechcad-kernel | mech_kernel/ai_orchestrator.py | .py | 568af85c83b58444 | 7 | 0 |
"""
MechKernel Capability Registry
v2 升级基础设施：自动注册 + schema 描述 + 参数校验 + 权限分级
专家第 5 轮审查建议：
- 当前 PUBLIC_OPS = frozenset({...}) 硬编码白名单不可维护
- 真实 LLM 需要结构化 op 描述
- 需要权限分级（public / read / internal）
- 需要参数类型/范围/必填校验
设计目标：
- 装饰器自动注册（@cap.register(...)）
- JSON Schema 风格的 input_schema
- LLM 友好的 list_public() 输出
"""
from typing... | vanyu0710/mechcad-kernel | mech_kernel/capability_registry.py | .py | f8dabb3c6cd220c7 | 7 | 0 |
"""
MechKernel 类型化错误定义（v1.1 修复版）
5 类 + 1 类：
1. InvalidRequestError - 编程错误，必须抛
2. KernelBugError - 内部 bug，必须抛
3. StateCorruptionError - 状态损坏，必须抛
4. GEOMETRY_FAILURE - 预期内几何失败，StepResult 表达
5. RECOVERABLE - 可修复失败，StepResult 表达
6. NOT_IMPLEMENTED - 能力未实现（占位 API），StepResult 表达（P0 修复新增）... | vanyu0710/mechcad-kernel | mech_kernel/errors.py | .py | cdb2bd3ff75cd3be | 7 | 0 |
"""
Demo 3: M1 阶段 — 用 MockMesh 展示 3D 渲染
M1 阶段实现：
- geometry_inspector：BRep 指标（体积/面数/包围盒/流形/水密/连通）
- renderer：matplotlib 离屏 4 视角渲染
- adaptive_renderer：智能决定何时渲染
"""
import os
import sys
sys.path.insert(0, '/workspace')
from mech_kernel import MechKernel
class MockBox:
"""Mock 一个 10x10x10 的立方体"""
def __init__(... | vanyu0710/mechcad-kernel | mech_kernel/examples/03_mock_render.py | .py | efa394bb1feb145f | 7 | 0 |
"""
Demo 4: M2 E2E — 用户说"建一个圆柱体"，AI 自动建
展示 MechAgent v1 单 orchestrator：
- Mock Planner 解析用户 prompt
- Mock Vision 验证
- Kernel 一步一步执行
- 每步打印结果
"""
import sys
import os
sys.path.insert(0, '/workspace')
from mech_kernel import MechKernel
from mech_kernel.ai_orchestrator import MockPlanner, MockVision, run_loop, PlannerAc... | vanyu0710/mechcad-kernel | mech_kernel/examples/04_e2e_orchestrator.py | .py | 1148849bbeb2bea3 | 7 | 0 |
"""
Demo 06: 端到端 Vision LLM + Planner LLM + MechKernel + 真实 build123d 几何
流程：
1. 准备手绘 PNG（matplotlib 模拟）
2. Vision LLM（DeepSeek Vision）→ 零件 JSON
3. Planner LLM（DeepSeek Chat）→ op 序列
4. MechKernel 执行 op → build123d 真实几何
5. 渲染对比图（手绘草图 vs 真实几何）
运行：PYTHONPATH=/workspace python3 mech_kernel/examples/06_end_to_end_llm.py
环境... | vanyu0710/mechcad-kernel | mech_kernel/examples/06_end_to_end_llm.py | .py | 8fd8ad5897eafa46 | 7 | 0 |
"""
Demo 07: 复杂实际零件端到端
4 个工程件：
1. 带孔方板（80×60×10 + Ø20 通孔）
2. 阶梯轴（Ø40×30 + Ø25×50，两段同轴）
3. 键槽轴（Ø40×80 + 12×5×40 键槽）
4. L 形支架（两个矩形 extrude union）
每件：
- 画手绘 PNG
- Vision LLM 识别
- Planner LLM 拆 op
- Kernel 真实几何
- 渲染对比图
"""
from __future__ import annotations
import os
import sys
import json
import base64
from pathlib impo... | vanyu0710/mechcad-kernel | mech_kernel/examples/07_complex_parts.py | .py | 5311fc44924ab0ac | 7 | 0 |
"""
Demo 09: Fillet + Chamfer 在真实零件上的应用
4 个 demo:
1. 圆角立方体（基本 fillet）
2. 倒角方板（基本 chamfer）
3. 带沉孔 + 圆角（fillet 复合 boolean）
4. 法兰盘带圆角（fillet 圆盘边缘 + 沉孔边）
每个 demo 输出对比图（带 fillet/chamfer 前 vs 后）
"""
from __future__ import annotations
import os, sys, math
from pathlib import Path
HERE = Path(__file__).parent
OUT = HERE / "f... | vanyu0710/mechcad-kernel | mech_kernel/examples/09_fillet_chamfer.py | .py | 3351ece2401f86d2 | 7 | 0 |
"""
Demo 10: Boolean op 显式 API（v1.7 新增）
4 个 boolean demo:
1. union: 两个 box 合并
2. subtract: 盒 - 圆柱（多 tool 一起切）
3. intersect: 盒 ∩ 圆柱
4. 实际工程：L 形盒用 boolean union
"""
from __future__ import annotations
import os, math
from pathlib import Path
HERE = Path(__file__).parent
OUT = HERE / "boolean_out"
OUT.mkdir(exist_ok=True)... | vanyu0710/mechcad-kernel | mech_kernel/examples/10_boolean.py | .py | a8a74aeb5f1b6968 | 7 | 0 |
"""
Demo 11: Hole + Mirror + Linear Pattern（v1.8-1.10 新增）
4 个工程件:
1. 法兰盘：4 孔用 hole
2. 对称键槽：用 mirror
3. 散热板：8 孔用 linear_pattern
4. 综合：板 + 4 hole + mirror + fillet（完整流程）
"""
from __future__ import annotations
import math
from pathlib import Path
HERE = Path(__file__).parent
OUT = HERE / "hole_out"
OUT.mkdir(exist_ok=Tru... | vanyu0710/mechcad-kernel | mech_kernel/examples/11_hole_mirror_pattern.py | .py | 79bd84e7cf44d570 | 7 | 0 |
"""
Demo 12: Query / Select / Measure (v1.11-1.15)
5 个 query 能力 + select 按类型 + measure 3 种度量 + delete/update
"""
from __future__ import annotations
from pathlib import Path
HERE = Path(__file__).parent
OUT = HERE / "query_out"
OUT.mkdir(exist_ok=True)
def make_bracket():
"""造一个测试件：带孔的板"""
from mech_kernel im... | vanyu0710/mechcad-kernel | mech_kernel/examples/12_query_measure.py | .py | 55e36318e0ce9fc0 | 7 | 0 |
"""
MechKernel Feature Graph（DAG）
替代 v1.0 的 List[FeatureNode]，支持：
- 拓扑排序
- 循环检测
- 依赖追踪
- 子图删除
"""
from typing import Dict, List, Set, Optional
from collections import deque
from .features import FeatureNode, FeatureState
from .errors import StateCorruptionError, KernelBugError
class FeatureGraph:
"""
Featur... | vanyu0710/mechcad-kernel | mech_kernel/feature_graph.py | .py | 698d48a129004efe | 7 | 0 |
"""
MechKernel Geometry Inspector（M1.1 修复版）
P0-2 修复：manifold/watertight/connected 用三态 (valid/invalid/unknown)
避免欧拉公式误判（圆环 g=1 欧拉=0，不能用 V-E+F==2 判断）
其他修复：
- bbox 解析容错（NaN/Inf → 兜底）
- 异常隔离，绝不让 kernel 崩
"""
from typing import Any, Optional, Tuple, Literal
import math
from .step_result import GeometrySummary
# 拓扑检查结果：... | vanyu0710/mechcad-kernel | mech_kernel/geometry_inspector.py | .py | 3738f6a2cfcbcf31 | 7.5 | 0 |
"""
MechKernel LLM 客户端（DeepSeek OpenAI 兼容）
专家 v8 审查后接真实 LLM（替换 MockPlanner / MockVision）。
DeepSeek 视觉模型：deepseek-v4-flash-vision-exp
DeepSeek 推理模型：deepseek-reasoner
DeepSeek 通用模型：deepseek-chat
环境变量：DSKEY（DeepSeek 官方 API key）
"""
from __future__ import annotations
import os
import json
import logging
import base64
imp... | vanyu0710/mechcad-kernel | mech_kernel/llm/deepseek.py | .py | 375f96dbc896713f | 7 | 0 |
"""
MechKernel Persistent Naming（v1.1 修复版）
P1-4 修复：
- 改用 semantic_name -> role -> candidates[] 结构
- 查询必须显式 role
- 多候选返回歧义结果（不静默选最新）
专家审查原话：
"如果查询只按 semantic_name，body 和 top_face 会产生歧义，
最新时间戳可能导致非确定性覆盖。建议改成 semantic_name -> role -> candidates[]，
查询必须显式要求 role；多候选时返回 ambiguity，而不是静默选最新。"
"""
from typing import Dict, Li... | vanyu0710/mechcad-kernel | mech_kernel/persistent_naming.py | .py | 35edcd1fe7a81ce4 | 7 | 0 |
"""
MechKernel Renderer（M1.1 修复版）
第 4 轮专家审查修复：
- P0-1 缓存键：用 (id, geometry_revision, level, config) 替代裸 id
- P0-4 异常隔离：坏几何/NaN/空/缺顶点不崩
- LRU 限制（默认 32 个）
- 任何异常都隔离，**绝不**让 kernel 崩溃
"""
from typing import Any, Optional, Dict, List, Tuple
from collections import OrderedDict
import io
import math
try:
import matplotl... | vanyu0710/mechcad-kernel | mech_kernel/renderer.py | .py | ce75fce12dbd3555 | 7 | 0 |
"""
MechKernel StepResult 数据结构
C 方案：自适应渲染 - 默认不渲染，拓扑变化 / 关键决策点才渲染
"""
from dataclasses import dataclass, field
from typing import List, Optional, Dict, Any, Literal
import time
import base64
RenderLevel = Literal["none", "iso_only", "full"]
ErrorKind = Literal["INVALID_REQUEST", "GEOMETRY_FAILURE", "RECOVERABLE", "KE... | vanyu0710/mechcad-kernel | mech_kernel/step_result.py | .py | 079ef5bd62b2a3ee | 7 | 0 |
"""
测试 5 类类型化错误
"""
import pytest
from mech_kernel.errors import (
MechKernelError, InvalidRequestError, KernelBugError, StateCorruptionError,
make_geometry_failure, make_recoverable, GeometryFailureReason
)
def test_invalid_request_error_inherits_base():
"""InvalidRequestError 继承 MechKernelError"""
e... | vanyu0710/mechcad-kernel | mech_kernel/tests/test_errors.py | .py | e4d8fe11edfe8ce7 | 7.5 | 0 |
"""
测试 M1：GeometryInspector + Renderer + AdaptiveRenderer
"""
import pytest
from mech_kernel.geometry_inspector import GeometryInspector
from mech_kernel.renderer import Renderer
from mech_kernel.adaptive_renderer import AdaptiveRenderer
from mech_kernel import MechKernel
from mech_kernel.step_result import GeometrySum... | vanyu0710/mechcad-kernel | mech_kernel/tests/test_m1.py | .py | b9338263d6b92f9d | 7.5 | 0 |
"""
测试 M2: AI Orchestrator（v1 单 Agent）
- Mock Planner 解析用户指令生成 plan
- run_loop 跑通 Plan→Execute→Inspect→Decide 循环
- E2E：建圆柱/立方体/法兰盘
"""
import pytest
from mech_kernel import MechKernel
from mech_kernel.ai_orchestrator import (
MockPlanner, MockVision, run_loop, PlannerAction
)
def test_planner_action_to_dict():
... | vanyu0710/mechcad-kernel | mech_kernel/tests/test_m2_orchestrator.py | .py | c16dbf4d2768e98c | 7.5 | 0 |
"""
测试 P0 修复：
1. NOT_IMPLEMENTED 错误类型
2. 事务 / undo 栈不被污染
3. 失败渲染策略（按 error_kind）
4. Reference 的 frozen=True
"""
import pytest
from mech_kernel import (
MechKernel, InvalidRequestError, KernelBugError, StateCorruptionError
)
from mech_kernel.errors import make_not_implemented, get_render_level_for_error
from mech_ke... | vanyu0710/mechcad-kernel | mech_kernel/tests/test_p0_fixes.py | .py | 8dce0685a164144d | 7.5 | 0 |
"""
测试 savepoint 模型（P0 事务修复 #2）
- 默认嵌套：内层 join 外层（只有一个 undo entry）
- savepoint 模式：内层独立入 undo
- __exit__ finally：rollback 异常不影响 _txn_depth 递减
"""
import pytest
from mech_kernel import MechKernel, InvalidRequestError, DeprecatedInternalAPIError
from mech_kernel.transaction import Transaction
def test_default_nested_jo... | vanyu0710/mechcad-kernel | mech_kernel/tests/test_savepoint.py | .py | 5e0f7927240e5940 | 7.5 | 0 |
"""
MechKernel 事务管理（v1.1 修复版 #2）
P3 原则：任何操作都在事务中，失败整体回滚。
P0 修复（专家第 3 轮）：
- `__exit__` 必须有 `finally` 保护 _txn_depth（rollback 抛异常时不递减）
- 删除 _push_undo 尸体 API（不再静默 pass）
- savepoint 明确语义：
- 默认嵌套：内层 join 外层（只在外层 commit 时入 undo）
- savepoint 显式 API：`Transaction.savepoint()` 标记内层独立点
P1 优化：
- 只在 commit 时才深拷贝完整状态（之前每次都拍两次... | vanyu0710/mechcad-kernel | mech_kernel/transaction.py | .py | f6b31ff0ca95cf2c | 7 | 0 |
"""
MechKernel 单位管理
P7 原则：内部全部毫米，API 不暴露 inch。
所有输入参数默认 mm，返回值默认 mm。
"""
from typing import Union
Number = Union[int, float]
# 内部单位
INTERNAL_UNIT = "mm"
# 单位转换表（输入侧只支持 mm 和 inch）
_TO_MM = {
"mm": 1.0,
"cm": 10.0,
"m": 1000.0,
"in": 25.4,
"inch": 25.4,
"ft": 304.8,
}
def to_mm(value: Number... | vanyu0710/mechcad-kernel | mech_kernel/units.py | .py | 1cd58141a059807e | 7 | 0 |
"""
MechKernel 输入校验器（v1.1 修复版）
P1-6 修复：统一行为
- 校验函数：返回规范化值
- 失败时：抛 InvalidRequestError
- 不再做"返回清洗值 vs 抛异常"的混合行为
专家审查原话：
"有的函数抛异常，有的返回清洗后的值，调用方很容易漏处理。
统一为'纯校验返回规范化值，失败统一抛 InvalidRequestError'，不要混合隐式修正。"
"""
from typing import Any, List, Tuple, Optional
from .errors import InvalidRequestError
from .units import is_posit... | vanyu0710/mechcad-kernel | mech_kernel/validators.py | .py | a0ead4027bd00fd6 | 7 | 0 |
"""Canonical fingerprinting for MiniMax H3 cached-CLIP requests.
Turns (prompt, the kwargs SpyClipProxy.tokenize() captured, clip identity,
cache schema version) into one deterministic sha256 hex digest, used as the
cache key. Pure function of the data already captured by the proxy -- no
ComfyUI imports, no GPU, no di... | Mu5hr00moO/ComfyUI-MiniMaxH3-CLIPCached | minimaxh3_clipcache/fingerprint.py | .py | 8bc13c325131bc27 | 7 | 0 |
"""Thin wrappers around ComfyUI's own CLIP-loading path (folder_paths,
comfy.sd.load_clip), used only for what CachedClipProxy needs: identifying
the encoder file on disk for the cache fingerprint, and lazily constructing
the real clip object on a cache MISS. Neither function reimplements
ComfyUI's loading logic -- bot... | Mu5hr00moO/ComfyUI-MiniMaxH3-CLIPCached | minimaxh3_clipcache/loader.py | .py | 2f7a284f8c7ba618 | 7 | 0 |
"""CachedClipProxy: transparent clip.tokenize()/encode_from_tokens_scheduled()
replacement that fingerprints the request and serves/saves conditioning from
disk instead of always re-running the real Qwen3-VL encoder.
tokenize() is lazy (phase 4 of the project plan): it does not touch
real_clip at all, it just remember... | Mu5hr00moO/ComfyUI-MiniMaxH3-CLIPCached | minimaxh3_clipcache/proxy.py | .py | dcc1080924f6320f | 7 | 0 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.