mcptoolshop-rnd 1.1.2.0.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- mcptoolshop_rnd-1.1.2.0.0/LICENSE +21 -0
- mcptoolshop_rnd-1.1.2.0.0/PKG-INFO +213 -0
- mcptoolshop_rnd-1.1.2.0.0/README.md +187 -0
- mcptoolshop_rnd-1.1.2.0.0/mcptoolshop_rnd.egg-info/PKG-INFO +213 -0
- mcptoolshop_rnd-1.1.2.0.0/mcptoolshop_rnd.egg-info/SOURCES.txt +19 -0
- mcptoolshop_rnd-1.1.2.0.0/mcptoolshop_rnd.egg-info/dependency_links.txt +1 -0
- mcptoolshop_rnd-1.1.2.0.0/mcptoolshop_rnd.egg-info/entry_points.txt +2 -0
- mcptoolshop_rnd-1.1.2.0.0/mcptoolshop_rnd.egg-info/top_level.txt +1 -0
- mcptoolshop_rnd-1.1.2.0.0/pyproject.toml +42 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/__init__.py +3 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/__main__.py +5 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/catalog.py +169 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/cli.py +464 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/frontmatter.py +196 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/model.py +227 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/readouts.py +82 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/release.py +123 -0
- mcptoolshop_rnd-1.1.2.0.0/rnd/store.py +164 -0
- mcptoolshop_rnd-1.1.2.0.0/setup.cfg +4 -0
- mcptoolshop_rnd-1.1.2.0.0/tests/test_cli.py +405 -0
- mcptoolshop_rnd-1.1.2.0.0/tests/test_rnd.py +310 -0
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 mcp-tool-shop
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE.
|
|
@@ -0,0 +1,213 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: mcptoolshop-rnd
|
|
3
|
+
Version: 1.1.2.0.0
|
|
4
|
+
Summary: The studio's research bench: a Markdown research library with a rebuildable SQLite index, driven by the rnd command.
|
|
5
|
+
Author-email: mcp-tool-shop <64996768+mcp-tool-shop@users.noreply.github.com>
|
|
6
|
+
License-Expression: MIT
|
|
7
|
+
Project-URL: Homepage, https://mcp-tool-shop-org.github.io/rnd/
|
|
8
|
+
Project-URL: Repository, https://github.com/mcp-tool-shop-org/rnd
|
|
9
|
+
Project-URL: Changelog, https://github.com/mcp-tool-shop-org/rnd/blob/main/CHANGELOG.md
|
|
10
|
+
Project-URL: Issues, https://github.com/mcp-tool-shop-org/rnd/issues
|
|
11
|
+
Keywords: research,knowledge-base,markdown,sqlite,fts5,cli
|
|
12
|
+
Classifier: Development Status :: 5 - Production/Stable
|
|
13
|
+
Classifier: Environment :: Console
|
|
14
|
+
Classifier: Intended Audience :: Developers
|
|
15
|
+
Classifier: Intended Audience :: Science/Research
|
|
16
|
+
Classifier: Operating System :: OS Independent
|
|
17
|
+
Classifier: Programming Language :: Python :: 3
|
|
18
|
+
Classifier: Programming Language :: Python :: 3.10
|
|
19
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
20
|
+
Classifier: Topic :: Scientific/Engineering
|
|
21
|
+
Classifier: Topic :: Text Processing :: Markup :: Markdown
|
|
22
|
+
Requires-Python: >=3.10
|
|
23
|
+
Description-Content-Type: text/markdown
|
|
24
|
+
License-File: LICENSE
|
|
25
|
+
Dynamic: license-file
|
|
26
|
+
|
|
27
|
+
<p align="center">
|
|
28
|
+
<a href="README.md">English</a> | <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
|
|
29
|
+
</p>
|
|
30
|
+
|
|
31
|
+
<p align="center">
|
|
32
|
+
<img src="https://raw.githubusercontent.com/mcp-tool-shop-org/brand/main/logos/rnd/readme.png" alt="Research and Development" width="400">
|
|
33
|
+
</p>
|
|
34
|
+
|
|
35
|
+
<p align="center">
|
|
36
|
+
<a href="https://github.com/mcp-tool-shop-org/rnd/actions/workflows/ci.yml"><img src="https://github.com/mcp-tool-shop-org/rnd/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
|
|
37
|
+
<a href="https://codecov.io/gh/mcp-tool-shop-org/rnd"><img src="https://codecov.io/gh/mcp-tool-shop-org/rnd/branch/main/graph/badge.svg" alt="Coverage"></a>
|
|
38
|
+
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="MIT License"></a>
|
|
39
|
+
<a href="https://mcp-tool-shop-org.github.io/rnd/"><img src="https://img.shields.io/badge/Landing_Page-live-blue" alt="Landing Page"></a>
|
|
40
|
+
</p>
|
|
41
|
+
|
|
42
|
+
The studio's research bench. Findings from any field come in fast as short
|
|
43
|
+
Markdown entries, with every source tagged and every claim marked as checked or
|
|
44
|
+
not. Experiments sit next to the entries they test. External catalogues are
|
|
45
|
+
mirrored and rated, and a registry lists the studio tools research can call on.
|
|
46
|
+
One command searches all of it.
|
|
47
|
+
|
|
48
|
+
## Where it sits: the bench before readouts
|
|
49
|
+
|
|
50
|
+
Two stores hold the studio's knowledge, and they do different jobs.
|
|
51
|
+
|
|
52
|
+
| | Research and Development (this repo) | [readouts](https://github.com/mcp-tool-shop-org/readouts) |
|
|
53
|
+
|---|---|---|
|
|
54
|
+
| Role | the bench: intake, experiments, open questions | the shelf: verified knowledge bases |
|
|
55
|
+
| Pace | an entry in minutes, claims start `unverified` | built and checked by study swarms |
|
|
56
|
+
| Shape | Markdown entries, one topic each | one SQLite knowledge base per domain |
|
|
57
|
+
| How messy | messy on purpose: disputed claims, dead ends and interim results stay visible | not messy at all: every row is sourced and verified |
|
|
58
|
+
|
|
59
|
+
Knowledge moves one way:
|
|
60
|
+
|
|
61
|
+
1. **Bench.** A finding lands here as an entry. Its claims start `[unverified]`,
|
|
62
|
+
and are marked `[verified]` only with a note of what checked them.
|
|
63
|
+
Measurements made on our own machines go in as `rig` sources, with their
|
|
64
|
+
harness under `experiments/`.
|
|
65
|
+
2. **Internal shelf.** Once a topic's load-bearing claims hold up, it is built
|
|
66
|
+
into a knowledge base in readouts' private working repo, or added to one.
|
|
67
|
+
3. **Public shelf.** A knowledge base is published to the public readouts repo
|
|
68
|
+
when it is added to the export allow-list.
|
|
69
|
+
|
|
70
|
+
`rnd readouts` searches every readouts knowledge base from here, so one seat
|
|
71
|
+
reaches both stores. Search the bench first, the shelf second, then research the gap.
|
|
72
|
+
|
|
73
|
+
Research here is not limited to the studio's own fields. Each entry records two
|
|
74
|
+
things separately: what the knowledge is, and what it means for the studio
|
|
75
|
+
(`relevance: act | watch | reference`). `reference` is a fine verdict.
|
|
76
|
+
|
|
77
|
+
## Use it
|
|
78
|
+
|
|
79
|
+
Requires Python 3.10 or later and nothing else (standard library only). Runs on
|
|
80
|
+
Windows, macOS and Linux. From the repo root:
|
|
81
|
+
|
|
82
|
+
```bash
|
|
83
|
+
python -m rnd search cuda graphs # full-text over entries + catalogues
|
|
84
|
+
python -m rnd show 2026-10-07-cuda-graphs # one entry, with backlinks
|
|
85
|
+
python -m rnd list --relevance act # what needs doing
|
|
86
|
+
python -m rnd tools # instruments this seat can use
|
|
87
|
+
python -m rnd readouts splice glitch --any # search the readouts knowledge bases too
|
|
88
|
+
python -m rnd catalog lanes # NVIDIA skills by lane, with studio fit
|
|
89
|
+
python -m rnd catalog list --fit adjacent # skills worth using when the need arises
|
|
90
|
+
python -m rnd new "Paper title" --kind paper --field audio --tag pitch
|
|
91
|
+
python -m rnd check # validate every file (exit 1 on errors)
|
|
92
|
+
python -m rnd sql "SELECT tier, count(*) FROM sources GROUP BY tier"
|
|
93
|
+
python -m rnd bump --note "what changed" # micro version bump + CHANGELOG section
|
|
94
|
+
```
|
|
95
|
+
|
|
96
|
+
Every listing command takes `--json` for agents. `rnd.cmd` (Windows) and
|
|
97
|
+
`rnd.sh` (POSIX shells) are thin wrappers, so the command works from any directory.
|
|
98
|
+
|
|
99
|
+
To use the tool from other projects, install it from PyPI:
|
|
100
|
+
|
|
101
|
+
```bash
|
|
102
|
+
pip install mcptoolshop-rnd
|
|
103
|
+
```
|
|
104
|
+
|
|
105
|
+
That installs the `rnd` command, not the library: the entries live in this
|
|
106
|
+
repository. The command finds a library in this order: `--library DIR`, then
|
|
107
|
+
`$RND_ROOT`, then the nearest folder at or above the current one that holds
|
|
108
|
+
`entries/` and `instruments/`. Outside a library it stops with `NO_LIBRARY`;
|
|
109
|
+
only `rnd readouts` works without one. The package imports as `rnd`, as does an
|
|
110
|
+
unrelated PyPI package called `rnd`, so don't install both in one environment.
|
|
111
|
+
|
|
112
|
+
The library changes daily, so versions have five segments,
|
|
113
|
+
`MAJOR.MINOR.PATCH.MICRO.NANO`. The first three version the `rnd` tool; MICRO marks a
|
|
114
|
+
structural library change and NANO an ordinary update. `rnd bump` raises the
|
|
115
|
+
last segment by default and writes a CHANGELOG section from the files changed
|
|
116
|
+
since the last tag, so each update gets its own small tagged version.
|
|
117
|
+
|
|
118
|
+
Exit codes: `0` ok · `1` invalid library files · `2` usage error or not found ·
|
|
119
|
+
`3` runtime failure (an external tool, or an unexpected error). Errors print a
|
|
120
|
+
code, a message and a hint; `--debug` adds the traceback.
|
|
121
|
+
|
|
122
|
+
## Layout
|
|
123
|
+
|
|
124
|
+
| Path | What | Who edits |
|
|
125
|
+
|------|------|-----------|
|
|
126
|
+
| `entries/YYYY/*.md` | Research entries: the source of truth | people and agents |
|
|
127
|
+
| `experiments/<name>/` | Harnesses, pinned inputs and result receipts for rig measurements | people and agents |
|
|
128
|
+
| `instruments/*.md` | Studio tools and protocols the seat can call (`kind: instrument`) | people and agents |
|
|
129
|
+
| `catalogs/<name>/source.json` | Where a catalogue comes from | people |
|
|
130
|
+
| `catalogs/<name>/catalog.json` | Pinned snapshot from `rnd catalog sync` | generated; never hand-edit |
|
|
131
|
+
| `catalogs/<name>/review.json` | Studio fit and notes per family and item | people |
|
|
132
|
+
| `rnd/` | The CLI | code |
|
|
133
|
+
| `rnd.db` | SQLite FTS5 index, rebuilt automatically when files change | generated; not in git |
|
|
134
|
+
|
|
135
|
+
## Entry format
|
|
136
|
+
|
|
137
|
+
```markdown
|
|
138
|
+
---
|
|
139
|
+
id: 2026-10-07-cuda-graphs # defaults to the file name
|
|
140
|
+
title: CUDA Graphs
|
|
141
|
+
date: 2026-10-07
|
|
142
|
+
kind: concept # finding concept release paper tool catalog rig-fact event question decision instrument
|
|
143
|
+
relevance: reference # act | watch | reference
|
|
144
|
+
fields: [gpu-computing] # any research field, open vocabulary
|
|
145
|
+
tags: [cuda-graphs, pytorch]
|
|
146
|
+
---
|
|
147
|
+
|
|
148
|
+
## Summary
|
|
149
|
+
## Key points
|
|
150
|
+
## Studio relevance
|
|
151
|
+
## Claims
|
|
152
|
+
- [unverified] A checkable statement.
|
|
153
|
+
- [verified] A checked statement. (via: what checked it, date)
|
|
154
|
+
## Sources
|
|
155
|
+
- [primary] https://… — publisher
|
|
156
|
+
```
|
|
157
|
+
|
|
158
|
+
- **Source tiers:** `primary` (vendor docs, papers, repos), `secondary`
|
|
159
|
+
(reputable write-ups), `aggregator` (summary sites, AI search output),
|
|
160
|
+
`user` (supplied by a person: slides, notes), `rig` (measured on our machines).
|
|
161
|
+
- **Claim confidence:** `unverified`, `verified`, `disputed`, `wrong`. A
|
|
162
|
+
`verified` or `wrong` claim must say what checked it with `(via: …)`.
|
|
163
|
+
- `[[entry-id]]` links entries; `rnd show` lists backlinks.
|
|
164
|
+
|
|
165
|
+
## Relation to other studio tools
|
|
166
|
+
|
|
167
|
+
- **readouts** is the verified shelf this bench feeds (see above).
|
|
168
|
+
- **research-os** builds a gated, frozen evidence pack for one topic. A topic a
|
|
169
|
+
decision depends on can graduate into a research-os pack, linked from the entry.
|
|
170
|
+
- **repo-knowledge** indexes the studio's own repos; this library covers
|
|
171
|
+
knowledge from outside them.
|
|
172
|
+
- Findings gathered while designing something belong here too, so they outlive
|
|
173
|
+
the session that produced them.
|
|
174
|
+
|
|
175
|
+
See `python -m rnd tools` for the full instrument registry, and
|
|
176
|
+
[docs/standards.md](docs/standards.md) for how the workflow scores against the
|
|
177
|
+
studio's workflow standards.
|
|
178
|
+
|
|
179
|
+
## Security and trust
|
|
180
|
+
|
|
181
|
+
- **Data touched:** files inside this repo (`entries/`, `instruments/`,
|
|
182
|
+
`catalogs/`, `experiments/`) and the index `rnd.db`, which it rebuilds. `rnd
|
|
183
|
+
readouts` opens the readouts knowledge bases **read-only**. `rnd sql` runs
|
|
184
|
+
against a read-only connection.
|
|
185
|
+
- **Data not touched:** nothing outside the repo and the readouts checkout. It
|
|
186
|
+
stores no credentials and reads none.
|
|
187
|
+
- **Network:** none, except `rnd catalog sync`, which calls the GitHub API
|
|
188
|
+
through your own `gh` CLI login when you run it.
|
|
189
|
+
- **Permissions:** ordinary file access. No elevated rights, no background service.
|
|
190
|
+
- **No telemetry.** Nothing is collected or sent.
|
|
191
|
+
- **Public repo hygiene:** before every push, the tree is scanned for
|
|
192
|
+
home-directory paths and operator identity.
|
|
193
|
+
|
|
194
|
+
Report vulnerabilities as described in [SECURITY.md](SECURITY.md).
|
|
195
|
+
|
|
196
|
+
## Tests
|
|
197
|
+
|
|
198
|
+
```bash
|
|
199
|
+
bash verify.sh # tests, library check, index build, smoke
|
|
200
|
+
python -m unittest discover -s tests -t . # tests only
|
|
201
|
+
```
|
|
202
|
+
|
|
203
|
+
## Status and licence
|
|
204
|
+
|
|
205
|
+
Maintained by the studio and in daily use. Code: [MIT](LICENSE). Entries: CC BY
|
|
206
|
+
4.0. The NVIDIA skills mirror in `catalogs/nvidia-skills/catalog.json`
|
|
207
|
+
reproduces skill names and descriptions from
|
|
208
|
+
[NVIDIA/skills](https://github.com/NVIDIA/skills) under that project's licences
|
|
209
|
+
(Apache-2.0 for code, CC-BY-4.0 for skill text).
|
|
210
|
+
|
|
211
|
+
---
|
|
212
|
+
|
|
213
|
+
<p align="center">Built by <a href="https://mcp-tool-shop.github.io/">MCP Tool Shop</a></p>
|
|
@@ -0,0 +1,187 @@
|
|
|
1
|
+
<p align="center">
|
|
2
|
+
<a href="README.md">English</a> | <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
|
|
3
|
+
</p>
|
|
4
|
+
|
|
5
|
+
<p align="center">
|
|
6
|
+
<img src="https://raw.githubusercontent.com/mcp-tool-shop-org/brand/main/logos/rnd/readme.png" alt="Research and Development" width="400">
|
|
7
|
+
</p>
|
|
8
|
+
|
|
9
|
+
<p align="center">
|
|
10
|
+
<a href="https://github.com/mcp-tool-shop-org/rnd/actions/workflows/ci.yml"><img src="https://github.com/mcp-tool-shop-org/rnd/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
|
|
11
|
+
<a href="https://codecov.io/gh/mcp-tool-shop-org/rnd"><img src="https://codecov.io/gh/mcp-tool-shop-org/rnd/branch/main/graph/badge.svg" alt="Coverage"></a>
|
|
12
|
+
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="MIT License"></a>
|
|
13
|
+
<a href="https://mcp-tool-shop-org.github.io/rnd/"><img src="https://img.shields.io/badge/Landing_Page-live-blue" alt="Landing Page"></a>
|
|
14
|
+
</p>
|
|
15
|
+
|
|
16
|
+
The studio's research bench. Findings from any field come in fast as short
|
|
17
|
+
Markdown entries, with every source tagged and every claim marked as checked or
|
|
18
|
+
not. Experiments sit next to the entries they test. External catalogues are
|
|
19
|
+
mirrored and rated, and a registry lists the studio tools research can call on.
|
|
20
|
+
One command searches all of it.
|
|
21
|
+
|
|
22
|
+
## Where it sits: the bench before readouts
|
|
23
|
+
|
|
24
|
+
Two stores hold the studio's knowledge, and they do different jobs.
|
|
25
|
+
|
|
26
|
+
| | Research and Development (this repo) | [readouts](https://github.com/mcp-tool-shop-org/readouts) |
|
|
27
|
+
|---|---|---|
|
|
28
|
+
| Role | the bench: intake, experiments, open questions | the shelf: verified knowledge bases |
|
|
29
|
+
| Pace | an entry in minutes, claims start `unverified` | built and checked by study swarms |
|
|
30
|
+
| Shape | Markdown entries, one topic each | one SQLite knowledge base per domain |
|
|
31
|
+
| How messy | messy on purpose: disputed claims, dead ends and interim results stay visible | not messy at all: every row is sourced and verified |
|
|
32
|
+
|
|
33
|
+
Knowledge moves one way:
|
|
34
|
+
|
|
35
|
+
1. **Bench.** A finding lands here as an entry. Its claims start `[unverified]`,
|
|
36
|
+
and are marked `[verified]` only with a note of what checked them.
|
|
37
|
+
Measurements made on our own machines go in as `rig` sources, with their
|
|
38
|
+
harness under `experiments/`.
|
|
39
|
+
2. **Internal shelf.** Once a topic's load-bearing claims hold up, it is built
|
|
40
|
+
into a knowledge base in readouts' private working repo, or added to one.
|
|
41
|
+
3. **Public shelf.** A knowledge base is published to the public readouts repo
|
|
42
|
+
when it is added to the export allow-list.
|
|
43
|
+
|
|
44
|
+
`rnd readouts` searches every readouts knowledge base from here, so one seat
|
|
45
|
+
reaches both stores. Search the bench first, the shelf second, then research the gap.
|
|
46
|
+
|
|
47
|
+
Research here is not limited to the studio's own fields. Each entry records two
|
|
48
|
+
things separately: what the knowledge is, and what it means for the studio
|
|
49
|
+
(`relevance: act | watch | reference`). `reference` is a fine verdict.
|
|
50
|
+
|
|
51
|
+
## Use it
|
|
52
|
+
|
|
53
|
+
Requires Python 3.10 or later and nothing else (standard library only). Runs on
|
|
54
|
+
Windows, macOS and Linux. From the repo root:
|
|
55
|
+
|
|
56
|
+
```bash
|
|
57
|
+
python -m rnd search cuda graphs # full-text over entries + catalogues
|
|
58
|
+
python -m rnd show 2026-10-07-cuda-graphs # one entry, with backlinks
|
|
59
|
+
python -m rnd list --relevance act # what needs doing
|
|
60
|
+
python -m rnd tools # instruments this seat can use
|
|
61
|
+
python -m rnd readouts splice glitch --any # search the readouts knowledge bases too
|
|
62
|
+
python -m rnd catalog lanes # NVIDIA skills by lane, with studio fit
|
|
63
|
+
python -m rnd catalog list --fit adjacent # skills worth using when the need arises
|
|
64
|
+
python -m rnd new "Paper title" --kind paper --field audio --tag pitch
|
|
65
|
+
python -m rnd check # validate every file (exit 1 on errors)
|
|
66
|
+
python -m rnd sql "SELECT tier, count(*) FROM sources GROUP BY tier"
|
|
67
|
+
python -m rnd bump --note "what changed" # micro version bump + CHANGELOG section
|
|
68
|
+
```
|
|
69
|
+
|
|
70
|
+
Every listing command takes `--json` for agents. `rnd.cmd` (Windows) and
|
|
71
|
+
`rnd.sh` (POSIX shells) are thin wrappers, so the command works from any directory.
|
|
72
|
+
|
|
73
|
+
To use the tool from other projects, install it from PyPI:
|
|
74
|
+
|
|
75
|
+
```bash
|
|
76
|
+
pip install mcptoolshop-rnd
|
|
77
|
+
```
|
|
78
|
+
|
|
79
|
+
That installs the `rnd` command, not the library: the entries live in this
|
|
80
|
+
repository. The command finds a library in this order: `--library DIR`, then
|
|
81
|
+
`$RND_ROOT`, then the nearest folder at or above the current one that holds
|
|
82
|
+
`entries/` and `instruments/`. Outside a library it stops with `NO_LIBRARY`;
|
|
83
|
+
only `rnd readouts` works without one. The package imports as `rnd`, as does an
|
|
84
|
+
unrelated PyPI package called `rnd`, so don't install both in one environment.
|
|
85
|
+
|
|
86
|
+
The library changes daily, so versions have five segments,
|
|
87
|
+
`MAJOR.MINOR.PATCH.MICRO.NANO`. The first three version the `rnd` tool; MICRO marks a
|
|
88
|
+
structural library change and NANO an ordinary update. `rnd bump` raises the
|
|
89
|
+
last segment by default and writes a CHANGELOG section from the files changed
|
|
90
|
+
since the last tag, so each update gets its own small tagged version.
|
|
91
|
+
|
|
92
|
+
Exit codes: `0` ok · `1` invalid library files · `2` usage error or not found ·
|
|
93
|
+
`3` runtime failure (an external tool, or an unexpected error). Errors print a
|
|
94
|
+
code, a message and a hint; `--debug` adds the traceback.
|
|
95
|
+
|
|
96
|
+
## Layout
|
|
97
|
+
|
|
98
|
+
| Path | What | Who edits |
|
|
99
|
+
|------|------|-----------|
|
|
100
|
+
| `entries/YYYY/*.md` | Research entries: the source of truth | people and agents |
|
|
101
|
+
| `experiments/<name>/` | Harnesses, pinned inputs and result receipts for rig measurements | people and agents |
|
|
102
|
+
| `instruments/*.md` | Studio tools and protocols the seat can call (`kind: instrument`) | people and agents |
|
|
103
|
+
| `catalogs/<name>/source.json` | Where a catalogue comes from | people |
|
|
104
|
+
| `catalogs/<name>/catalog.json` | Pinned snapshot from `rnd catalog sync` | generated; never hand-edit |
|
|
105
|
+
| `catalogs/<name>/review.json` | Studio fit and notes per family and item | people |
|
|
106
|
+
| `rnd/` | The CLI | code |
|
|
107
|
+
| `rnd.db` | SQLite FTS5 index, rebuilt automatically when files change | generated; not in git |
|
|
108
|
+
|
|
109
|
+
## Entry format
|
|
110
|
+
|
|
111
|
+
```markdown
|
|
112
|
+
---
|
|
113
|
+
id: 2026-10-07-cuda-graphs # defaults to the file name
|
|
114
|
+
title: CUDA Graphs
|
|
115
|
+
date: 2026-10-07
|
|
116
|
+
kind: concept # finding concept release paper tool catalog rig-fact event question decision instrument
|
|
117
|
+
relevance: reference # act | watch | reference
|
|
118
|
+
fields: [gpu-computing] # any research field, open vocabulary
|
|
119
|
+
tags: [cuda-graphs, pytorch]
|
|
120
|
+
---
|
|
121
|
+
|
|
122
|
+
## Summary
|
|
123
|
+
## Key points
|
|
124
|
+
## Studio relevance
|
|
125
|
+
## Claims
|
|
126
|
+
- [unverified] A checkable statement.
|
|
127
|
+
- [verified] A checked statement. (via: what checked it, date)
|
|
128
|
+
## Sources
|
|
129
|
+
- [primary] https://… — publisher
|
|
130
|
+
```
|
|
131
|
+
|
|
132
|
+
- **Source tiers:** `primary` (vendor docs, papers, repos), `secondary`
|
|
133
|
+
(reputable write-ups), `aggregator` (summary sites, AI search output),
|
|
134
|
+
`user` (supplied by a person: slides, notes), `rig` (measured on our machines).
|
|
135
|
+
- **Claim confidence:** `unverified`, `verified`, `disputed`, `wrong`. A
|
|
136
|
+
`verified` or `wrong` claim must say what checked it with `(via: …)`.
|
|
137
|
+
- `[[entry-id]]` links entries; `rnd show` lists backlinks.
|
|
138
|
+
|
|
139
|
+
## Relation to other studio tools
|
|
140
|
+
|
|
141
|
+
- **readouts** is the verified shelf this bench feeds (see above).
|
|
142
|
+
- **research-os** builds a gated, frozen evidence pack for one topic. A topic a
|
|
143
|
+
decision depends on can graduate into a research-os pack, linked from the entry.
|
|
144
|
+
- **repo-knowledge** indexes the studio's own repos; this library covers
|
|
145
|
+
knowledge from outside them.
|
|
146
|
+
- Findings gathered while designing something belong here too, so they outlive
|
|
147
|
+
the session that produced them.
|
|
148
|
+
|
|
149
|
+
See `python -m rnd tools` for the full instrument registry, and
|
|
150
|
+
[docs/standards.md](docs/standards.md) for how the workflow scores against the
|
|
151
|
+
studio's workflow standards.
|
|
152
|
+
|
|
153
|
+
## Security and trust
|
|
154
|
+
|
|
155
|
+
- **Data touched:** files inside this repo (`entries/`, `instruments/`,
|
|
156
|
+
`catalogs/`, `experiments/`) and the index `rnd.db`, which it rebuilds. `rnd
|
|
157
|
+
readouts` opens the readouts knowledge bases **read-only**. `rnd sql` runs
|
|
158
|
+
against a read-only connection.
|
|
159
|
+
- **Data not touched:** nothing outside the repo and the readouts checkout. It
|
|
160
|
+
stores no credentials and reads none.
|
|
161
|
+
- **Network:** none, except `rnd catalog sync`, which calls the GitHub API
|
|
162
|
+
through your own `gh` CLI login when you run it.
|
|
163
|
+
- **Permissions:** ordinary file access. No elevated rights, no background service.
|
|
164
|
+
- **No telemetry.** Nothing is collected or sent.
|
|
165
|
+
- **Public repo hygiene:** before every push, the tree is scanned for
|
|
166
|
+
home-directory paths and operator identity.
|
|
167
|
+
|
|
168
|
+
Report vulnerabilities as described in [SECURITY.md](SECURITY.md).
|
|
169
|
+
|
|
170
|
+
## Tests
|
|
171
|
+
|
|
172
|
+
```bash
|
|
173
|
+
bash verify.sh # tests, library check, index build, smoke
|
|
174
|
+
python -m unittest discover -s tests -t . # tests only
|
|
175
|
+
```
|
|
176
|
+
|
|
177
|
+
## Status and licence
|
|
178
|
+
|
|
179
|
+
Maintained by the studio and in daily use. Code: [MIT](LICENSE). Entries: CC BY
|
|
180
|
+
4.0. The NVIDIA skills mirror in `catalogs/nvidia-skills/catalog.json`
|
|
181
|
+
reproduces skill names and descriptions from
|
|
182
|
+
[NVIDIA/skills](https://github.com/NVIDIA/skills) under that project's licences
|
|
183
|
+
(Apache-2.0 for code, CC-BY-4.0 for skill text).
|
|
184
|
+
|
|
185
|
+
---
|
|
186
|
+
|
|
187
|
+
<p align="center">Built by <a href="https://mcp-tool-shop.github.io/">MCP Tool Shop</a></p>
|
|
@@ -0,0 +1,213 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: mcptoolshop-rnd
|
|
3
|
+
Version: 1.1.2.0.0
|
|
4
|
+
Summary: The studio's research bench: a Markdown research library with a rebuildable SQLite index, driven by the rnd command.
|
|
5
|
+
Author-email: mcp-tool-shop <64996768+mcp-tool-shop@users.noreply.github.com>
|
|
6
|
+
License-Expression: MIT
|
|
7
|
+
Project-URL: Homepage, https://mcp-tool-shop-org.github.io/rnd/
|
|
8
|
+
Project-URL: Repository, https://github.com/mcp-tool-shop-org/rnd
|
|
9
|
+
Project-URL: Changelog, https://github.com/mcp-tool-shop-org/rnd/blob/main/CHANGELOG.md
|
|
10
|
+
Project-URL: Issues, https://github.com/mcp-tool-shop-org/rnd/issues
|
|
11
|
+
Keywords: research,knowledge-base,markdown,sqlite,fts5,cli
|
|
12
|
+
Classifier: Development Status :: 5 - Production/Stable
|
|
13
|
+
Classifier: Environment :: Console
|
|
14
|
+
Classifier: Intended Audience :: Developers
|
|
15
|
+
Classifier: Intended Audience :: Science/Research
|
|
16
|
+
Classifier: Operating System :: OS Independent
|
|
17
|
+
Classifier: Programming Language :: Python :: 3
|
|
18
|
+
Classifier: Programming Language :: Python :: 3.10
|
|
19
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
20
|
+
Classifier: Topic :: Scientific/Engineering
|
|
21
|
+
Classifier: Topic :: Text Processing :: Markup :: Markdown
|
|
22
|
+
Requires-Python: >=3.10
|
|
23
|
+
Description-Content-Type: text/markdown
|
|
24
|
+
License-File: LICENSE
|
|
25
|
+
Dynamic: license-file
|
|
26
|
+
|
|
27
|
+
<p align="center">
|
|
28
|
+
<a href="README.md">English</a> | <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
|
|
29
|
+
</p>
|
|
30
|
+
|
|
31
|
+
<p align="center">
|
|
32
|
+
<img src="https://raw.githubusercontent.com/mcp-tool-shop-org/brand/main/logos/rnd/readme.png" alt="Research and Development" width="400">
|
|
33
|
+
</p>
|
|
34
|
+
|
|
35
|
+
<p align="center">
|
|
36
|
+
<a href="https://github.com/mcp-tool-shop-org/rnd/actions/workflows/ci.yml"><img src="https://github.com/mcp-tool-shop-org/rnd/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
|
|
37
|
+
<a href="https://codecov.io/gh/mcp-tool-shop-org/rnd"><img src="https://codecov.io/gh/mcp-tool-shop-org/rnd/branch/main/graph/badge.svg" alt="Coverage"></a>
|
|
38
|
+
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="MIT License"></a>
|
|
39
|
+
<a href="https://mcp-tool-shop-org.github.io/rnd/"><img src="https://img.shields.io/badge/Landing_Page-live-blue" alt="Landing Page"></a>
|
|
40
|
+
</p>
|
|
41
|
+
|
|
42
|
+
The studio's research bench. Findings from any field come in fast as short
|
|
43
|
+
Markdown entries, with every source tagged and every claim marked as checked or
|
|
44
|
+
not. Experiments sit next to the entries they test. External catalogues are
|
|
45
|
+
mirrored and rated, and a registry lists the studio tools research can call on.
|
|
46
|
+
One command searches all of it.
|
|
47
|
+
|
|
48
|
+
## Where it sits: the bench before readouts
|
|
49
|
+
|
|
50
|
+
Two stores hold the studio's knowledge, and they do different jobs.
|
|
51
|
+
|
|
52
|
+
| | Research and Development (this repo) | [readouts](https://github.com/mcp-tool-shop-org/readouts) |
|
|
53
|
+
|---|---|---|
|
|
54
|
+
| Role | the bench: intake, experiments, open questions | the shelf: verified knowledge bases |
|
|
55
|
+
| Pace | an entry in minutes, claims start `unverified` | built and checked by study swarms |
|
|
56
|
+
| Shape | Markdown entries, one topic each | one SQLite knowledge base per domain |
|
|
57
|
+
| How messy | messy on purpose: disputed claims, dead ends and interim results stay visible | not messy at all: every row is sourced and verified |
|
|
58
|
+
|
|
59
|
+
Knowledge moves one way:
|
|
60
|
+
|
|
61
|
+
1. **Bench.** A finding lands here as an entry. Its claims start `[unverified]`,
|
|
62
|
+
and are marked `[verified]` only with a note of what checked them.
|
|
63
|
+
Measurements made on our own machines go in as `rig` sources, with their
|
|
64
|
+
harness under `experiments/`.
|
|
65
|
+
2. **Internal shelf.** Once a topic's load-bearing claims hold up, it is built
|
|
66
|
+
into a knowledge base in readouts' private working repo, or added to one.
|
|
67
|
+
3. **Public shelf.** A knowledge base is published to the public readouts repo
|
|
68
|
+
when it is added to the export allow-list.
|
|
69
|
+
|
|
70
|
+
`rnd readouts` searches every readouts knowledge base from here, so one seat
|
|
71
|
+
reaches both stores. Search the bench first, the shelf second, then research the gap.
|
|
72
|
+
|
|
73
|
+
Research here is not limited to the studio's own fields. Each entry records two
|
|
74
|
+
things separately: what the knowledge is, and what it means for the studio
|
|
75
|
+
(`relevance: act | watch | reference`). `reference` is a fine verdict.
|
|
76
|
+
|
|
77
|
+
## Use it
|
|
78
|
+
|
|
79
|
+
Requires Python 3.10 or later and nothing else (standard library only). Runs on
|
|
80
|
+
Windows, macOS and Linux. From the repo root:
|
|
81
|
+
|
|
82
|
+
```bash
|
|
83
|
+
python -m rnd search cuda graphs # full-text over entries + catalogues
|
|
84
|
+
python -m rnd show 2026-10-07-cuda-graphs # one entry, with backlinks
|
|
85
|
+
python -m rnd list --relevance act # what needs doing
|
|
86
|
+
python -m rnd tools # instruments this seat can use
|
|
87
|
+
python -m rnd readouts splice glitch --any # search the readouts knowledge bases too
|
|
88
|
+
python -m rnd catalog lanes # NVIDIA skills by lane, with studio fit
|
|
89
|
+
python -m rnd catalog list --fit adjacent # skills worth using when the need arises
|
|
90
|
+
python -m rnd new "Paper title" --kind paper --field audio --tag pitch
|
|
91
|
+
python -m rnd check # validate every file (exit 1 on errors)
|
|
92
|
+
python -m rnd sql "SELECT tier, count(*) FROM sources GROUP BY tier"
|
|
93
|
+
python -m rnd bump --note "what changed" # micro version bump + CHANGELOG section
|
|
94
|
+
```
|
|
95
|
+
|
|
96
|
+
Every listing command takes `--json` for agents. `rnd.cmd` (Windows) and
|
|
97
|
+
`rnd.sh` (POSIX shells) are thin wrappers, so the command works from any directory.
|
|
98
|
+
|
|
99
|
+
To use the tool from other projects, install it from PyPI:
|
|
100
|
+
|
|
101
|
+
```bash
|
|
102
|
+
pip install mcptoolshop-rnd
|
|
103
|
+
```
|
|
104
|
+
|
|
105
|
+
That installs the `rnd` command, not the library: the entries live in this
|
|
106
|
+
repository. The command finds a library in this order: `--library DIR`, then
|
|
107
|
+
`$RND_ROOT`, then the nearest folder at or above the current one that holds
|
|
108
|
+
`entries/` and `instruments/`. Outside a library it stops with `NO_LIBRARY`;
|
|
109
|
+
only `rnd readouts` works without one. The package imports as `rnd`, as does an
|
|
110
|
+
unrelated PyPI package called `rnd`, so don't install both in one environment.
|
|
111
|
+
|
|
112
|
+
The library changes daily, so versions have five segments,
|
|
113
|
+
`MAJOR.MINOR.PATCH.MICRO.NANO`. The first three version the `rnd` tool; MICRO marks a
|
|
114
|
+
structural library change and NANO an ordinary update. `rnd bump` raises the
|
|
115
|
+
last segment by default and writes a CHANGELOG section from the files changed
|
|
116
|
+
since the last tag, so each update gets its own small tagged version.
|
|
117
|
+
|
|
118
|
+
Exit codes: `0` ok · `1` invalid library files · `2` usage error or not found ·
|
|
119
|
+
`3` runtime failure (an external tool, or an unexpected error). Errors print a
|
|
120
|
+
code, a message and a hint; `--debug` adds the traceback.
|
|
121
|
+
|
|
122
|
+
## Layout
|
|
123
|
+
|
|
124
|
+
| Path | What | Who edits |
|
|
125
|
+
|------|------|-----------|
|
|
126
|
+
| `entries/YYYY/*.md` | Research entries: the source of truth | people and agents |
|
|
127
|
+
| `experiments/<name>/` | Harnesses, pinned inputs and result receipts for rig measurements | people and agents |
|
|
128
|
+
| `instruments/*.md` | Studio tools and protocols the seat can call (`kind: instrument`) | people and agents |
|
|
129
|
+
| `catalogs/<name>/source.json` | Where a catalogue comes from | people |
|
|
130
|
+
| `catalogs/<name>/catalog.json` | Pinned snapshot from `rnd catalog sync` | generated; never hand-edit |
|
|
131
|
+
| `catalogs/<name>/review.json` | Studio fit and notes per family and item | people |
|
|
132
|
+
| `rnd/` | The CLI | code |
|
|
133
|
+
| `rnd.db` | SQLite FTS5 index, rebuilt automatically when files change | generated; not in git |
|
|
134
|
+
|
|
135
|
+
## Entry format
|
|
136
|
+
|
|
137
|
+
```markdown
|
|
138
|
+
---
|
|
139
|
+
id: 2026-10-07-cuda-graphs # defaults to the file name
|
|
140
|
+
title: CUDA Graphs
|
|
141
|
+
date: 2026-10-07
|
|
142
|
+
kind: concept # finding concept release paper tool catalog rig-fact event question decision instrument
|
|
143
|
+
relevance: reference # act | watch | reference
|
|
144
|
+
fields: [gpu-computing] # any research field, open vocabulary
|
|
145
|
+
tags: [cuda-graphs, pytorch]
|
|
146
|
+
---
|
|
147
|
+
|
|
148
|
+
## Summary
|
|
149
|
+
## Key points
|
|
150
|
+
## Studio relevance
|
|
151
|
+
## Claims
|
|
152
|
+
- [unverified] A checkable statement.
|
|
153
|
+
- [verified] A checked statement. (via: what checked it, date)
|
|
154
|
+
## Sources
|
|
155
|
+
- [primary] https://… — publisher
|
|
156
|
+
```
|
|
157
|
+
|
|
158
|
+
- **Source tiers:** `primary` (vendor docs, papers, repos), `secondary`
|
|
159
|
+
(reputable write-ups), `aggregator` (summary sites, AI search output),
|
|
160
|
+
`user` (supplied by a person: slides, notes), `rig` (measured on our machines).
|
|
161
|
+
- **Claim confidence:** `unverified`, `verified`, `disputed`, `wrong`. A
|
|
162
|
+
`verified` or `wrong` claim must say what checked it with `(via: …)`.
|
|
163
|
+
- `[[entry-id]]` links entries; `rnd show` lists backlinks.
|
|
164
|
+
|
|
165
|
+
## Relation to other studio tools
|
|
166
|
+
|
|
167
|
+
- **readouts** is the verified shelf this bench feeds (see above).
|
|
168
|
+
- **research-os** builds a gated, frozen evidence pack for one topic. A topic a
|
|
169
|
+
decision depends on can graduate into a research-os pack, linked from the entry.
|
|
170
|
+
- **repo-knowledge** indexes the studio's own repos; this library covers
|
|
171
|
+
knowledge from outside them.
|
|
172
|
+
- Findings gathered while designing something belong here too, so they outlive
|
|
173
|
+
the session that produced them.
|
|
174
|
+
|
|
175
|
+
See `python -m rnd tools` for the full instrument registry, and
|
|
176
|
+
[docs/standards.md](docs/standards.md) for how the workflow scores against the
|
|
177
|
+
studio's workflow standards.
|
|
178
|
+
|
|
179
|
+
## Security and trust
|
|
180
|
+
|
|
181
|
+
- **Data touched:** files inside this repo (`entries/`, `instruments/`,
|
|
182
|
+
`catalogs/`, `experiments/`) and the index `rnd.db`, which it rebuilds. `rnd
|
|
183
|
+
readouts` opens the readouts knowledge bases **read-only**. `rnd sql` runs
|
|
184
|
+
against a read-only connection.
|
|
185
|
+
- **Data not touched:** nothing outside the repo and the readouts checkout. It
|
|
186
|
+
stores no credentials and reads none.
|
|
187
|
+
- **Network:** none, except `rnd catalog sync`, which calls the GitHub API
|
|
188
|
+
through your own `gh` CLI login when you run it.
|
|
189
|
+
- **Permissions:** ordinary file access. No elevated rights, no background service.
|
|
190
|
+
- **No telemetry.** Nothing is collected or sent.
|
|
191
|
+
- **Public repo hygiene:** before every push, the tree is scanned for
|
|
192
|
+
home-directory paths and operator identity.
|
|
193
|
+
|
|
194
|
+
Report vulnerabilities as described in [SECURITY.md](SECURITY.md).
|
|
195
|
+
|
|
196
|
+
## Tests
|
|
197
|
+
|
|
198
|
+
```bash
|
|
199
|
+
bash verify.sh # tests, library check, index build, smoke
|
|
200
|
+
python -m unittest discover -s tests -t . # tests only
|
|
201
|
+
```
|
|
202
|
+
|
|
203
|
+
## Status and licence
|
|
204
|
+
|
|
205
|
+
Maintained by the studio and in daily use. Code: [MIT](LICENSE). Entries: CC BY
|
|
206
|
+
4.0. The NVIDIA skills mirror in `catalogs/nvidia-skills/catalog.json`
|
|
207
|
+
reproduces skill names and descriptions from
|
|
208
|
+
[NVIDIA/skills](https://github.com/NVIDIA/skills) under that project's licences
|
|
209
|
+
(Apache-2.0 for code, CC-BY-4.0 for skill text).
|
|
210
|
+
|
|
211
|
+
---
|
|
212
|
+
|
|
213
|
+
<p align="center">Built by <a href="https://mcp-tool-shop.github.io/">MCP Tool Shop</a></p>
|
|
@@ -0,0 +1,19 @@
|
|
|
1
|
+
LICENSE
|
|
2
|
+
README.md
|
|
3
|
+
pyproject.toml
|
|
4
|
+
mcptoolshop_rnd.egg-info/PKG-INFO
|
|
5
|
+
mcptoolshop_rnd.egg-info/SOURCES.txt
|
|
6
|
+
mcptoolshop_rnd.egg-info/dependency_links.txt
|
|
7
|
+
mcptoolshop_rnd.egg-info/entry_points.txt
|
|
8
|
+
mcptoolshop_rnd.egg-info/top_level.txt
|
|
9
|
+
rnd/__init__.py
|
|
10
|
+
rnd/__main__.py
|
|
11
|
+
rnd/catalog.py
|
|
12
|
+
rnd/cli.py
|
|
13
|
+
rnd/frontmatter.py
|
|
14
|
+
rnd/model.py
|
|
15
|
+
rnd/readouts.py
|
|
16
|
+
rnd/release.py
|
|
17
|
+
rnd/store.py
|
|
18
|
+
tests/test_cli.py
|
|
19
|
+
tests/test_rnd.py
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
rnd
|