skillcheckup 0.3.4__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.
- skillcheckup-0.3.4/LICENSE +21 -0
- skillcheckup-0.3.4/PKG-INFO +147 -0
- skillcheckup-0.3.4/README.md +137 -0
- skillcheckup-0.3.4/pyproject.toml +17 -0
- skillcheckup-0.3.4/setup.cfg +4 -0
- skillcheckup-0.3.4/skillcheckup.egg-info/PKG-INFO +147 -0
- skillcheckup-0.3.4/skillcheckup.egg-info/SOURCES.txt +10 -0
- skillcheckup-0.3.4/skillcheckup.egg-info/dependency_links.txt +1 -0
- skillcheckup-0.3.4/skillcheckup.egg-info/entry_points.txt +2 -0
- skillcheckup-0.3.4/skillcheckup.egg-info/top_level.txt +1 -0
- skillcheckup-0.3.4/skilldoctor.py +638 -0
- skillcheckup-0.3.4/tests/test_skilldoctor.py +234 -0
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 Bui Van Dong
|
|
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 +1,147 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: skillcheckup
|
|
3
|
+
Version: 0.3.4
|
|
4
|
+
Summary: Audit the AI skill files installed on your machine for validity, security risks, conflicts, and token cost.
|
|
5
|
+
License: MIT
|
|
6
|
+
Requires-Python: >=3.10
|
|
7
|
+
Description-Content-Type: text/markdown
|
|
8
|
+
License-File: LICENSE
|
|
9
|
+
Dynamic: license-file
|
|
10
|
+
|
|
11
|
+
<p align="center">
|
|
12
|
+
<img src="assets/masthead.png" alt="SkillDoctor: a checkup for the skills your AI runs" width="880">
|
|
13
|
+
</p>
|
|
14
|
+
|
|
15
|
+
<p align="center">
|
|
16
|
+
<a href="https://github.com/Dongdarealest/skilldoctor/releases"><img src="https://img.shields.io/github/v/release/Dongdarealest/skilldoctor?style=flat-square&color=141414&label=release" alt="Latest release"></a>
|
|
17
|
+
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-141414?style=flat-square" alt="MIT license"></a>
|
|
18
|
+
<img src="https://img.shields.io/badge/python-3.10%2B-141414?style=flat-square" alt="Python 3.10 or newer">
|
|
19
|
+
<img src="https://img.shields.io/badge/dependencies-zero-141414?style=flat-square" alt="Zero dependencies">
|
|
20
|
+
</p>
|
|
21
|
+
|
|
22
|
+
Skills are instructions your AI loads and follows, often at the start of every conversation. Most people install them from repositories, gists, and chat messages without ever reading them. SkillDoctor reads them for you: one command audits every skill on your machine and gives each one a verdict.
|
|
23
|
+
|
|
24
|
+
<p align="center">
|
|
25
|
+
<img src="assets/demo.png" alt="SkillDoctor demo output: one clean skill passes, an invalid skill and a risky skill fail with reasons" width="880">
|
|
26
|
+
</p>
|
|
27
|
+
|
|
28
|
+
<p align="center"><em>The bundled demo: one clean skill passes, two fail, every failure with its reason.</em></p>
|
|
29
|
+
|
|
30
|
+
**The State of Public Skills:** we audited 777 skills from the six biggest public collections. One in three failed. [Read the audit](AUDIT.md).
|
|
31
|
+
|
|
32
|
+
## Quick start
|
|
33
|
+
|
|
34
|
+
With Python already set up, the familiar way works too:
|
|
35
|
+
|
|
36
|
+
```bash
|
|
37
|
+
pip install skillcheckup
|
|
38
|
+
```
|
|
39
|
+
|
|
40
|
+
One line installs a `skilldoctor` command on your machine:
|
|
41
|
+
|
|
42
|
+
```bash
|
|
43
|
+
curl -fsSL https://raw.githubusercontent.com/Dongdarealest/skilldoctor/main/install.sh | bash
|
|
44
|
+
```
|
|
45
|
+
|
|
46
|
+
On Windows, in PowerShell:
|
|
47
|
+
|
|
48
|
+
```powershell
|
|
49
|
+
irm https://raw.githubusercontent.com/Dongdarealest/skilldoctor/main/install.ps1 | iex
|
|
50
|
+
```
|
|
51
|
+
|
|
52
|
+
Then it is just:
|
|
53
|
+
|
|
54
|
+
```bash
|
|
55
|
+
# Audit the skill folders on this machine
|
|
56
|
+
skilldoctor
|
|
57
|
+
|
|
58
|
+
# See it work in ten seconds on the bundled examples
|
|
59
|
+
skilldoctor --demo
|
|
60
|
+
|
|
61
|
+
# Audit a specific folder
|
|
62
|
+
skilldoctor ~/path/to/skills
|
|
63
|
+
|
|
64
|
+
# Machine readable output for scripts and CI
|
|
65
|
+
skilldoctor --json
|
|
66
|
+
```
|
|
67
|
+
|
|
68
|
+
No installer fan? The whole tool is one file: download `skilldoctor.py` from this repo and run `python3 skilldoctor.py` the same way. It needs Python 3.10 or newer and nothing else.
|
|
69
|
+
|
|
70
|
+
## What it checks
|
|
71
|
+
|
|
72
|
+
| Check | What it catches |
|
|
73
|
+
| --- | --- |
|
|
74
|
+
| Validity | Frontmatter that does not parse, missing name or description, names that are not kebab-case or do not match their folder |
|
|
75
|
+
| Security | Downloads piped into a shell, instructions touching credential stores, language that sends user data out, prompt override phrases, uninspectable base64 blobs, destructive commands |
|
|
76
|
+
| Conflicts | The same skill name installed in several host folders, near-duplicate descriptions that make hosts load the wrong skill |
|
|
77
|
+
| Token cost | The estimated tokens each skill adds, and the total always-on cost loaded into every conversation |
|
|
78
|
+
|
|
79
|
+
## Reading the report
|
|
80
|
+
|
|
81
|
+
- **OK**: the skill parses, identifies itself properly, and shows no risky patterns.
|
|
82
|
+
- **WARN**: worth a human look, for example a reference to a credential store or a very long file.
|
|
83
|
+
- **FAIL**: broken or dangerous. The report exits with code 1 when any skill fails, so it drops straight into CI.
|
|
84
|
+
|
|
85
|
+
## Scores and fixes
|
|
86
|
+
|
|
87
|
+
Every skill gets a score out of 100 and a letter grade, and so does the whole installation. Errors and high risk findings cost the most, warnings cost a little, and a clean skill sits at 100. Most findings also carry a fix line that says exactly what to change, so the report reads as a work list, not just a verdict.
|
|
88
|
+
|
|
89
|
+
## Removing what fails
|
|
90
|
+
|
|
91
|
+
Finding a bad skill is only half the job, so SkillDoctor can clear it out too. Removal is deliberately boring and reversible:
|
|
92
|
+
|
|
93
|
+
```bash
|
|
94
|
+
# Audit, then move every FAIL skill out of its host folder
|
|
95
|
+
skilldoctor --quarantine-failed
|
|
96
|
+
|
|
97
|
+
# The same, with no confirmation prompt, for scripts
|
|
98
|
+
skilldoctor --quarantine-failed --yes
|
|
99
|
+
|
|
100
|
+
# Changed your mind? Put the last batch back
|
|
101
|
+
skilldoctor --restore
|
|
102
|
+
```
|
|
103
|
+
|
|
104
|
+
Nothing is deleted outright. Failed skills move to `~/.skilldoctor/quarantine` in a dated batch with a manifest of where each one came from, and `--restore` puts the most recent batch back exactly where it was. Only FAIL verdicts are moved; a WARN is a judgment call, and judgment calls stay yours.
|
|
105
|
+
|
|
106
|
+
## Show your audit
|
|
107
|
+
|
|
108
|
+
Publish skills of your own? If your collection passes a SkillDoctor audit, say so where people can see it. This is the badge on a repo, linked back here:
|
|
109
|
+
|
|
110
|
+
[](https://github.com/Dongdarealest/skilldoctor)
|
|
111
|
+
|
|
112
|
+
Paste this into your README and it renders as the badge above:
|
|
113
|
+
|
|
114
|
+
```markdown
|
|
115
|
+
[](https://github.com/Dongdarealest/skilldoctor)
|
|
116
|
+
```
|
|
117
|
+
|
|
118
|
+
## In CI, and as a GitHub Action
|
|
119
|
+
|
|
120
|
+
This repo doubles as a GitHub Action, so a collection can audit itself on every push:
|
|
121
|
+
|
|
122
|
+
```yaml
|
|
123
|
+
- uses: Dongdarealest/skilldoctor@v0.3.4
|
|
124
|
+
```
|
|
125
|
+
|
|
126
|
+
It fails the build when any skill fails the audit. Anywhere else, the plain command does the same job in CI: `skilldoctor --json` for machines, exit code 1 on any FAIL.
|
|
127
|
+
|
|
128
|
+
## Why this exists
|
|
129
|
+
|
|
130
|
+
Public skill collections are growing faster than anyone can review them, and audits of popular collections keep finding files that are invalid, duplicated, or carry instructions no user would accept if they read them. Installing a skill is trusting a stranger's instructions inside your own AI. A checkup should take ten seconds, so here it is.
|
|
131
|
+
|
|
132
|
+
## Support
|
|
133
|
+
|
|
134
|
+
SkillDoctor is free and MIT licensed. If it saved you from one bad install, a coffee keeps the next check coming.
|
|
135
|
+
|
|
136
|
+
<p>
|
|
137
|
+
<a href="https://github.com/sponsors/Dongdarealest"><img src="https://img.shields.io/badge/GitHub%20Sponsors-141414?style=for-the-badge&logo=githubsponsors&logoColor=EA4AAA" alt="Sponsor on GitHub"></a>
|
|
138
|
+
<a href="https://ko-fi.com/dongdarealest"><img src="https://img.shields.io/badge/Ko--fi-141414?style=for-the-badge&logo=kofi&logoColor=FF5E5B" alt="Buy me a coffee on Ko-fi"></a>
|
|
139
|
+
</p>
|
|
140
|
+
|
|
141
|
+
## Contributing
|
|
142
|
+
|
|
143
|
+
New checks are the most welcome contributions, especially patterns for risks this version does not catch yet. Open an issue with the skill that slipped through, or send a pull request with a test that proves the new check works.
|
|
144
|
+
|
|
145
|
+
## License
|
|
146
|
+
|
|
147
|
+
MIT. See [LICENSE](LICENSE).
|
|
@@ -0,0 +1,137 @@
|
|
|
1
|
+
<p align="center">
|
|
2
|
+
<img src="assets/masthead.png" alt="SkillDoctor: a checkup for the skills your AI runs" width="880">
|
|
3
|
+
</p>
|
|
4
|
+
|
|
5
|
+
<p align="center">
|
|
6
|
+
<a href="https://github.com/Dongdarealest/skilldoctor/releases"><img src="https://img.shields.io/github/v/release/Dongdarealest/skilldoctor?style=flat-square&color=141414&label=release" alt="Latest release"></a>
|
|
7
|
+
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-141414?style=flat-square" alt="MIT license"></a>
|
|
8
|
+
<img src="https://img.shields.io/badge/python-3.10%2B-141414?style=flat-square" alt="Python 3.10 or newer">
|
|
9
|
+
<img src="https://img.shields.io/badge/dependencies-zero-141414?style=flat-square" alt="Zero dependencies">
|
|
10
|
+
</p>
|
|
11
|
+
|
|
12
|
+
Skills are instructions your AI loads and follows, often at the start of every conversation. Most people install them from repositories, gists, and chat messages without ever reading them. SkillDoctor reads them for you: one command audits every skill on your machine and gives each one a verdict.
|
|
13
|
+
|
|
14
|
+
<p align="center">
|
|
15
|
+
<img src="assets/demo.png" alt="SkillDoctor demo output: one clean skill passes, an invalid skill and a risky skill fail with reasons" width="880">
|
|
16
|
+
</p>
|
|
17
|
+
|
|
18
|
+
<p align="center"><em>The bundled demo: one clean skill passes, two fail, every failure with its reason.</em></p>
|
|
19
|
+
|
|
20
|
+
**The State of Public Skills:** we audited 777 skills from the six biggest public collections. One in three failed. [Read the audit](AUDIT.md).
|
|
21
|
+
|
|
22
|
+
## Quick start
|
|
23
|
+
|
|
24
|
+
With Python already set up, the familiar way works too:
|
|
25
|
+
|
|
26
|
+
```bash
|
|
27
|
+
pip install skillcheckup
|
|
28
|
+
```
|
|
29
|
+
|
|
30
|
+
One line installs a `skilldoctor` command on your machine:
|
|
31
|
+
|
|
32
|
+
```bash
|
|
33
|
+
curl -fsSL https://raw.githubusercontent.com/Dongdarealest/skilldoctor/main/install.sh | bash
|
|
34
|
+
```
|
|
35
|
+
|
|
36
|
+
On Windows, in PowerShell:
|
|
37
|
+
|
|
38
|
+
```powershell
|
|
39
|
+
irm https://raw.githubusercontent.com/Dongdarealest/skilldoctor/main/install.ps1 | iex
|
|
40
|
+
```
|
|
41
|
+
|
|
42
|
+
Then it is just:
|
|
43
|
+
|
|
44
|
+
```bash
|
|
45
|
+
# Audit the skill folders on this machine
|
|
46
|
+
skilldoctor
|
|
47
|
+
|
|
48
|
+
# See it work in ten seconds on the bundled examples
|
|
49
|
+
skilldoctor --demo
|
|
50
|
+
|
|
51
|
+
# Audit a specific folder
|
|
52
|
+
skilldoctor ~/path/to/skills
|
|
53
|
+
|
|
54
|
+
# Machine readable output for scripts and CI
|
|
55
|
+
skilldoctor --json
|
|
56
|
+
```
|
|
57
|
+
|
|
58
|
+
No installer fan? The whole tool is one file: download `skilldoctor.py` from this repo and run `python3 skilldoctor.py` the same way. It needs Python 3.10 or newer and nothing else.
|
|
59
|
+
|
|
60
|
+
## What it checks
|
|
61
|
+
|
|
62
|
+
| Check | What it catches |
|
|
63
|
+
| --- | --- |
|
|
64
|
+
| Validity | Frontmatter that does not parse, missing name or description, names that are not kebab-case or do not match their folder |
|
|
65
|
+
| Security | Downloads piped into a shell, instructions touching credential stores, language that sends user data out, prompt override phrases, uninspectable base64 blobs, destructive commands |
|
|
66
|
+
| Conflicts | The same skill name installed in several host folders, near-duplicate descriptions that make hosts load the wrong skill |
|
|
67
|
+
| Token cost | The estimated tokens each skill adds, and the total always-on cost loaded into every conversation |
|
|
68
|
+
|
|
69
|
+
## Reading the report
|
|
70
|
+
|
|
71
|
+
- **OK**: the skill parses, identifies itself properly, and shows no risky patterns.
|
|
72
|
+
- **WARN**: worth a human look, for example a reference to a credential store or a very long file.
|
|
73
|
+
- **FAIL**: broken or dangerous. The report exits with code 1 when any skill fails, so it drops straight into CI.
|
|
74
|
+
|
|
75
|
+
## Scores and fixes
|
|
76
|
+
|
|
77
|
+
Every skill gets a score out of 100 and a letter grade, and so does the whole installation. Errors and high risk findings cost the most, warnings cost a little, and a clean skill sits at 100. Most findings also carry a fix line that says exactly what to change, so the report reads as a work list, not just a verdict.
|
|
78
|
+
|
|
79
|
+
## Removing what fails
|
|
80
|
+
|
|
81
|
+
Finding a bad skill is only half the job, so SkillDoctor can clear it out too. Removal is deliberately boring and reversible:
|
|
82
|
+
|
|
83
|
+
```bash
|
|
84
|
+
# Audit, then move every FAIL skill out of its host folder
|
|
85
|
+
skilldoctor --quarantine-failed
|
|
86
|
+
|
|
87
|
+
# The same, with no confirmation prompt, for scripts
|
|
88
|
+
skilldoctor --quarantine-failed --yes
|
|
89
|
+
|
|
90
|
+
# Changed your mind? Put the last batch back
|
|
91
|
+
skilldoctor --restore
|
|
92
|
+
```
|
|
93
|
+
|
|
94
|
+
Nothing is deleted outright. Failed skills move to `~/.skilldoctor/quarantine` in a dated batch with a manifest of where each one came from, and `--restore` puts the most recent batch back exactly where it was. Only FAIL verdicts are moved; a WARN is a judgment call, and judgment calls stay yours.
|
|
95
|
+
|
|
96
|
+
## Show your audit
|
|
97
|
+
|
|
98
|
+
Publish skills of your own? If your collection passes a SkillDoctor audit, say so where people can see it. This is the badge on a repo, linked back here:
|
|
99
|
+
|
|
100
|
+
[](https://github.com/Dongdarealest/skilldoctor)
|
|
101
|
+
|
|
102
|
+
Paste this into your README and it renders as the badge above:
|
|
103
|
+
|
|
104
|
+
```markdown
|
|
105
|
+
[](https://github.com/Dongdarealest/skilldoctor)
|
|
106
|
+
```
|
|
107
|
+
|
|
108
|
+
## In CI, and as a GitHub Action
|
|
109
|
+
|
|
110
|
+
This repo doubles as a GitHub Action, so a collection can audit itself on every push:
|
|
111
|
+
|
|
112
|
+
```yaml
|
|
113
|
+
- uses: Dongdarealest/skilldoctor@v0.3.4
|
|
114
|
+
```
|
|
115
|
+
|
|
116
|
+
It fails the build when any skill fails the audit. Anywhere else, the plain command does the same job in CI: `skilldoctor --json` for machines, exit code 1 on any FAIL.
|
|
117
|
+
|
|
118
|
+
## Why this exists
|
|
119
|
+
|
|
120
|
+
Public skill collections are growing faster than anyone can review them, and audits of popular collections keep finding files that are invalid, duplicated, or carry instructions no user would accept if they read them. Installing a skill is trusting a stranger's instructions inside your own AI. A checkup should take ten seconds, so here it is.
|
|
121
|
+
|
|
122
|
+
## Support
|
|
123
|
+
|
|
124
|
+
SkillDoctor is free and MIT licensed. If it saved you from one bad install, a coffee keeps the next check coming.
|
|
125
|
+
|
|
126
|
+
<p>
|
|
127
|
+
<a href="https://github.com/sponsors/Dongdarealest"><img src="https://img.shields.io/badge/GitHub%20Sponsors-141414?style=for-the-badge&logo=githubsponsors&logoColor=EA4AAA" alt="Sponsor on GitHub"></a>
|
|
128
|
+
<a href="https://ko-fi.com/dongdarealest"><img src="https://img.shields.io/badge/Ko--fi-141414?style=for-the-badge&logo=kofi&logoColor=FF5E5B" alt="Buy me a coffee on Ko-fi"></a>
|
|
129
|
+
</p>
|
|
130
|
+
|
|
131
|
+
## Contributing
|
|
132
|
+
|
|
133
|
+
New checks are the most welcome contributions, especially patterns for risks this version does not catch yet. Open an issue with the skill that slipped through, or send a pull request with a test that proves the new check works.
|
|
134
|
+
|
|
135
|
+
## License
|
|
136
|
+
|
|
137
|
+
MIT. See [LICENSE](LICENSE).
|
|
@@ -0,0 +1,17 @@
|
|
|
1
|
+
[build-system]
|
|
2
|
+
requires = ["setuptools>=61"]
|
|
3
|
+
build-backend = "setuptools.build_meta"
|
|
4
|
+
|
|
5
|
+
[project]
|
|
6
|
+
name = "skillcheckup"
|
|
7
|
+
version = "0.3.4"
|
|
8
|
+
description = "Audit the AI skill files installed on your machine for validity, security risks, conflicts, and token cost."
|
|
9
|
+
readme = "README.md"
|
|
10
|
+
license = { text = "MIT" }
|
|
11
|
+
requires-python = ">=3.10"
|
|
12
|
+
|
|
13
|
+
[project.scripts]
|
|
14
|
+
skilldoctor = "skilldoctor:main"
|
|
15
|
+
|
|
16
|
+
[tool.setuptools]
|
|
17
|
+
py-modules = ["skilldoctor"]
|
|
@@ -0,0 +1,147 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: skillcheckup
|
|
3
|
+
Version: 0.3.4
|
|
4
|
+
Summary: Audit the AI skill files installed on your machine for validity, security risks, conflicts, and token cost.
|
|
5
|
+
License: MIT
|
|
6
|
+
Requires-Python: >=3.10
|
|
7
|
+
Description-Content-Type: text/markdown
|
|
8
|
+
License-File: LICENSE
|
|
9
|
+
Dynamic: license-file
|
|
10
|
+
|
|
11
|
+
<p align="center">
|
|
12
|
+
<img src="assets/masthead.png" alt="SkillDoctor: a checkup for the skills your AI runs" width="880">
|
|
13
|
+
</p>
|
|
14
|
+
|
|
15
|
+
<p align="center">
|
|
16
|
+
<a href="https://github.com/Dongdarealest/skilldoctor/releases"><img src="https://img.shields.io/github/v/release/Dongdarealest/skilldoctor?style=flat-square&color=141414&label=release" alt="Latest release"></a>
|
|
17
|
+
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-141414?style=flat-square" alt="MIT license"></a>
|
|
18
|
+
<img src="https://img.shields.io/badge/python-3.10%2B-141414?style=flat-square" alt="Python 3.10 or newer">
|
|
19
|
+
<img src="https://img.shields.io/badge/dependencies-zero-141414?style=flat-square" alt="Zero dependencies">
|
|
20
|
+
</p>
|
|
21
|
+
|
|
22
|
+
Skills are instructions your AI loads and follows, often at the start of every conversation. Most people install them from repositories, gists, and chat messages without ever reading them. SkillDoctor reads them for you: one command audits every skill on your machine and gives each one a verdict.
|
|
23
|
+
|
|
24
|
+
<p align="center">
|
|
25
|
+
<img src="assets/demo.png" alt="SkillDoctor demo output: one clean skill passes, an invalid skill and a risky skill fail with reasons" width="880">
|
|
26
|
+
</p>
|
|
27
|
+
|
|
28
|
+
<p align="center"><em>The bundled demo: one clean skill passes, two fail, every failure with its reason.</em></p>
|
|
29
|
+
|
|
30
|
+
**The State of Public Skills:** we audited 777 skills from the six biggest public collections. One in three failed. [Read the audit](AUDIT.md).
|
|
31
|
+
|
|
32
|
+
## Quick start
|
|
33
|
+
|
|
34
|
+
With Python already set up, the familiar way works too:
|
|
35
|
+
|
|
36
|
+
```bash
|
|
37
|
+
pip install skillcheckup
|
|
38
|
+
```
|
|
39
|
+
|
|
40
|
+
One line installs a `skilldoctor` command on your machine:
|
|
41
|
+
|
|
42
|
+
```bash
|
|
43
|
+
curl -fsSL https://raw.githubusercontent.com/Dongdarealest/skilldoctor/main/install.sh | bash
|
|
44
|
+
```
|
|
45
|
+
|
|
46
|
+
On Windows, in PowerShell:
|
|
47
|
+
|
|
48
|
+
```powershell
|
|
49
|
+
irm https://raw.githubusercontent.com/Dongdarealest/skilldoctor/main/install.ps1 | iex
|
|
50
|
+
```
|
|
51
|
+
|
|
52
|
+
Then it is just:
|
|
53
|
+
|
|
54
|
+
```bash
|
|
55
|
+
# Audit the skill folders on this machine
|
|
56
|
+
skilldoctor
|
|
57
|
+
|
|
58
|
+
# See it work in ten seconds on the bundled examples
|
|
59
|
+
skilldoctor --demo
|
|
60
|
+
|
|
61
|
+
# Audit a specific folder
|
|
62
|
+
skilldoctor ~/path/to/skills
|
|
63
|
+
|
|
64
|
+
# Machine readable output for scripts and CI
|
|
65
|
+
skilldoctor --json
|
|
66
|
+
```
|
|
67
|
+
|
|
68
|
+
No installer fan? The whole tool is one file: download `skilldoctor.py` from this repo and run `python3 skilldoctor.py` the same way. It needs Python 3.10 or newer and nothing else.
|
|
69
|
+
|
|
70
|
+
## What it checks
|
|
71
|
+
|
|
72
|
+
| Check | What it catches |
|
|
73
|
+
| --- | --- |
|
|
74
|
+
| Validity | Frontmatter that does not parse, missing name or description, names that are not kebab-case or do not match their folder |
|
|
75
|
+
| Security | Downloads piped into a shell, instructions touching credential stores, language that sends user data out, prompt override phrases, uninspectable base64 blobs, destructive commands |
|
|
76
|
+
| Conflicts | The same skill name installed in several host folders, near-duplicate descriptions that make hosts load the wrong skill |
|
|
77
|
+
| Token cost | The estimated tokens each skill adds, and the total always-on cost loaded into every conversation |
|
|
78
|
+
|
|
79
|
+
## Reading the report
|
|
80
|
+
|
|
81
|
+
- **OK**: the skill parses, identifies itself properly, and shows no risky patterns.
|
|
82
|
+
- **WARN**: worth a human look, for example a reference to a credential store or a very long file.
|
|
83
|
+
- **FAIL**: broken or dangerous. The report exits with code 1 when any skill fails, so it drops straight into CI.
|
|
84
|
+
|
|
85
|
+
## Scores and fixes
|
|
86
|
+
|
|
87
|
+
Every skill gets a score out of 100 and a letter grade, and so does the whole installation. Errors and high risk findings cost the most, warnings cost a little, and a clean skill sits at 100. Most findings also carry a fix line that says exactly what to change, so the report reads as a work list, not just a verdict.
|
|
88
|
+
|
|
89
|
+
## Removing what fails
|
|
90
|
+
|
|
91
|
+
Finding a bad skill is only half the job, so SkillDoctor can clear it out too. Removal is deliberately boring and reversible:
|
|
92
|
+
|
|
93
|
+
```bash
|
|
94
|
+
# Audit, then move every FAIL skill out of its host folder
|
|
95
|
+
skilldoctor --quarantine-failed
|
|
96
|
+
|
|
97
|
+
# The same, with no confirmation prompt, for scripts
|
|
98
|
+
skilldoctor --quarantine-failed --yes
|
|
99
|
+
|
|
100
|
+
# Changed your mind? Put the last batch back
|
|
101
|
+
skilldoctor --restore
|
|
102
|
+
```
|
|
103
|
+
|
|
104
|
+
Nothing is deleted outright. Failed skills move to `~/.skilldoctor/quarantine` in a dated batch with a manifest of where each one came from, and `--restore` puts the most recent batch back exactly where it was. Only FAIL verdicts are moved; a WARN is a judgment call, and judgment calls stay yours.
|
|
105
|
+
|
|
106
|
+
## Show your audit
|
|
107
|
+
|
|
108
|
+
Publish skills of your own? If your collection passes a SkillDoctor audit, say so where people can see it. This is the badge on a repo, linked back here:
|
|
109
|
+
|
|
110
|
+
[](https://github.com/Dongdarealest/skilldoctor)
|
|
111
|
+
|
|
112
|
+
Paste this into your README and it renders as the badge above:
|
|
113
|
+
|
|
114
|
+
```markdown
|
|
115
|
+
[](https://github.com/Dongdarealest/skilldoctor)
|
|
116
|
+
```
|
|
117
|
+
|
|
118
|
+
## In CI, and as a GitHub Action
|
|
119
|
+
|
|
120
|
+
This repo doubles as a GitHub Action, so a collection can audit itself on every push:
|
|
121
|
+
|
|
122
|
+
```yaml
|
|
123
|
+
- uses: Dongdarealest/skilldoctor@v0.3.4
|
|
124
|
+
```
|
|
125
|
+
|
|
126
|
+
It fails the build when any skill fails the audit. Anywhere else, the plain command does the same job in CI: `skilldoctor --json` for machines, exit code 1 on any FAIL.
|
|
127
|
+
|
|
128
|
+
## Why this exists
|
|
129
|
+
|
|
130
|
+
Public skill collections are growing faster than anyone can review them, and audits of popular collections keep finding files that are invalid, duplicated, or carry instructions no user would accept if they read them. Installing a skill is trusting a stranger's instructions inside your own AI. A checkup should take ten seconds, so here it is.
|
|
131
|
+
|
|
132
|
+
## Support
|
|
133
|
+
|
|
134
|
+
SkillDoctor is free and MIT licensed. If it saved you from one bad install, a coffee keeps the next check coming.
|
|
135
|
+
|
|
136
|
+
<p>
|
|
137
|
+
<a href="https://github.com/sponsors/Dongdarealest"><img src="https://img.shields.io/badge/GitHub%20Sponsors-141414?style=for-the-badge&logo=githubsponsors&logoColor=EA4AAA" alt="Sponsor on GitHub"></a>
|
|
138
|
+
<a href="https://ko-fi.com/dongdarealest"><img src="https://img.shields.io/badge/Ko--fi-141414?style=for-the-badge&logo=kofi&logoColor=FF5E5B" alt="Buy me a coffee on Ko-fi"></a>
|
|
139
|
+
</p>
|
|
140
|
+
|
|
141
|
+
## Contributing
|
|
142
|
+
|
|
143
|
+
New checks are the most welcome contributions, especially patterns for risks this version does not catch yet. Open an issue with the skill that slipped through, or send a pull request with a test that proves the new check works.
|
|
144
|
+
|
|
145
|
+
## License
|
|
146
|
+
|
|
147
|
+
MIT. See [LICENSE](LICENSE).
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
LICENSE
|
|
2
|
+
README.md
|
|
3
|
+
pyproject.toml
|
|
4
|
+
skilldoctor.py
|
|
5
|
+
skillcheckup.egg-info/PKG-INFO
|
|
6
|
+
skillcheckup.egg-info/SOURCES.txt
|
|
7
|
+
skillcheckup.egg-info/dependency_links.txt
|
|
8
|
+
skillcheckup.egg-info/entry_points.txt
|
|
9
|
+
skillcheckup.egg-info/top_level.txt
|
|
10
|
+
tests/test_skilldoctor.py
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
skilldoctor
|