@kensio/skills 1.13.1
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.
- package/README.md +48 -0
- package/bin/kensio-skills.mjs +213 -0
- package/package.json +39 -0
- package/skills/dynamodb-single-table/SKILL.md +329 -0
- package/skills/dynamodb-single-table/references/aws-guidance.md +183 -0
- package/skills/github-issue-drafting/SKILL.md +260 -0
- package/skills/isolated-testing-style/SKILL.md +293 -0
- package/skills/pangram-check/SKILL.md +125 -0
- package/skills/pangram-check/references/configuration.md +120 -0
- package/skills/pangram-check/references/reading-results.md +69 -0
- package/skills/pangram-check/scripts/pangram-check.mjs +807 -0
- package/skills/part-factory-test-data/SKILL.md +238 -0
- package/skills/skill-template/SKILL.md +126 -0
- package/skills/technical-prose-style/SKILL.md +356 -0
- package/skills/technical-prose-style/references/measurements.md +356 -0
- package/skills/technical-prose-style/scripts/prose-check.mjs +381 -0
- package/skills/yulin-aws-simulation/SKILL.md +324 -0
|
@@ -0,0 +1,238 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: part-factory-test-data
|
|
3
|
+
description: Build test data with @kensio/part-factory, keeping in the factory everything a test does not care about, passing dependencies at call time so factories stay independent, and choosing between StaticFactory, DynamicFactory, VariantFactory, MappedFactory and AsyncMappedFactory. Use when writing test fixtures or builders, when a test file is full of object literals, when a shared event, message or payload shape is being hand-written, when a required field is about to be made optional to ease test setup, and when tempted to wrap a factory in a helper function that applies overrides.
|
|
4
|
+
license: Apache-2.0
|
|
5
|
+
metadata:
|
|
6
|
+
version: "1.13.1"
|
|
7
|
+
---
|
|
8
|
+
|
|
9
|
+
# Building test data with Part Factory
|
|
10
|
+
|
|
11
|
+
[Part Factory](https://partfactory.dev/) (`@kensio/part-factory`) builds typed objects for tests. A
|
|
12
|
+
factory holds the defaults, and `make(overrides)` returns an object with the overrides applied down
|
|
13
|
+
through the nested structure.
|
|
14
|
+
|
|
15
|
+
The package README is the authority on the API. This skill covers what to put in a factory, where
|
|
16
|
+
factories should live, and which one to reach for.
|
|
17
|
+
|
|
18
|
+
It serves the `isolated-testing-style` skill. Factories are what make building state inside each
|
|
19
|
+
test cheap enough that nobody reaches for a shared fixture in the first place.
|
|
20
|
+
|
|
21
|
+
## Say only what the test is about
|
|
22
|
+
|
|
23
|
+
A factory defines every value the test ignores. The test can state only the values it does. That is
|
|
24
|
+
the whole point of one.
|
|
25
|
+
|
|
26
|
+
Without it, a test opens with twenty lines of construction and it is unclear which of them the
|
|
27
|
+
assertions actually depend on. With it, the lines that are there are the lines that matter:
|
|
28
|
+
|
|
29
|
+
```typescript
|
|
30
|
+
it("charges VAT on the order total", () => {
|
|
31
|
+
// Given an order whose lines add up to a total the assertion depends on.
|
|
32
|
+
const order = orderFactory.make({ lines: [{ price: 1000 }, { price: 2000 }] });
|
|
33
|
+
|
|
34
|
+
// When it is priced.
|
|
35
|
+
const priced = priceOrder(order);
|
|
36
|
+
|
|
37
|
+
// Then VAT is a fifth of that total.
|
|
38
|
+
expect(priced.vat).toBe(600);
|
|
39
|
+
});
|
|
40
|
+
```
|
|
41
|
+
|
|
42
|
+
The prices are written down because the assertion is arithmetic on them. The order id, the customer
|
|
43
|
+
id and the dates stay with the factory, because the test would read the same whatever they were. The
|
|
44
|
+
rule is that simple: **if an assertion depends on a value, put it in the test. Otherwise let the
|
|
45
|
+
factory supply it.**
|
|
46
|
+
|
|
47
|
+
This also removes a pressure that quietly damages production types. When constructing an object by
|
|
48
|
+
hand is painful, the tempting fix is to mark its required fields optional so tests can skip them,
|
|
49
|
+
weakening the type for every caller in order to serve the tests. A factory makes the setup cheap.
|
|
50
|
+
The type can go on saying what is actually required.
|
|
51
|
+
|
|
52
|
+
## Overrides are a deep partial
|
|
53
|
+
|
|
54
|
+
Overrides merge into the defaults, all the way down the nested structure. That is what makes "state
|
|
55
|
+
only what matters" possible. A test can set one field three levels deep and leave its siblings
|
|
56
|
+
alone.
|
|
57
|
+
|
|
58
|
+
Two consequences worth knowing:
|
|
59
|
+
|
|
60
|
+
- Arrays override by index, so `{ lines: [{ price: 1000 }] }` replaces the first default line and
|
|
61
|
+
leaves any others in place.
|
|
62
|
+
- An empty array leaves the defaults in place. To assert on emptiness, build the case some other way
|
|
63
|
+
rather than expecting `{ lines: [] }` to do it.
|
|
64
|
+
|
|
65
|
+
## Which factory
|
|
66
|
+
|
|
67
|
+
- **`StaticFactory`** when the defaults are fixed values. The simplest thing that works, and the
|
|
68
|
+
right default choice.
|
|
69
|
+
- **`DynamicFactory`** when the defaults have to be generated fresh for each object. Pair it with
|
|
70
|
+
[`@faker-js/faker`](https://fakerjs.dev/). This is what gives tests their isolation. A random
|
|
71
|
+
email or a UUID means two tests cannot collide, so tearing down is unnecessary.
|
|
72
|
+
- **`VariantFactory`** for a named variation of a base factory, when the variation is a concept the
|
|
73
|
+
tests talk about. `closedOfferFactory` reads better in ten tests than
|
|
74
|
+
`offerFactory.make({ closesAt: aMinuteAgo })` written ten times.
|
|
75
|
+
- **`MappedFactory`** when the output shape differs from the parts you want to override.
|
|
76
|
+
- **`AsyncMappedFactory`** when producing the value means awaiting something, such as inserting a
|
|
77
|
+
row or signing a payload.
|
|
78
|
+
|
|
79
|
+
```typescript
|
|
80
|
+
import { DynamicFactory } from "@kensio/part-factory";
|
|
81
|
+
import { faker } from "@faker-js/faker";
|
|
82
|
+
|
|
83
|
+
export const customerFactory = new DynamicFactory<Customer>(() => ({
|
|
84
|
+
id: faker.string.uuid(),
|
|
85
|
+
email: faker.internet.email(),
|
|
86
|
+
name: faker.person.fullName(),
|
|
87
|
+
}));
|
|
88
|
+
```
|
|
89
|
+
|
|
90
|
+
## Add a variant only for a shared meaning
|
|
91
|
+
|
|
92
|
+
A variant earns its name when several tests mean the same thing by it. One test that needs an
|
|
93
|
+
unusual value should write that value down inline, and leave the factories alone.
|
|
94
|
+
|
|
95
|
+
The failure mode is a directory of `cancelledOrderWithRefundAndNoAddressFactory` names, each used
|
|
96
|
+
once, where reading a test means going to find out what its factory actually sets. Writing the field
|
|
97
|
+
in the test is shorter and says more.
|
|
98
|
+
|
|
99
|
+
## Reach for MappedFactory only when the map is a real transformation
|
|
100
|
+
|
|
101
|
+
`MappedFactory` earns its place when the thing you want to override has a different shape from the
|
|
102
|
+
thing you want back. Good cases:
|
|
103
|
+
|
|
104
|
+
- Parts to an encoded form body: `{ email, password }` mapped to a
|
|
105
|
+
`application/x-www-form-urlencoded` string.
|
|
106
|
+
- Front matter parts to a file as written on disk: `{ title, tags, body }` mapped to the YAML block
|
|
107
|
+
plus the markdown underneath.
|
|
108
|
+
- Components to a formatted identifier: ARN parts mapped to the ARN string.
|
|
109
|
+
|
|
110
|
+
If the mapping function is copying fields across into an object of the same shape, you wanted a
|
|
111
|
+
`DynamicFactory`. An identity map adds a type parameter, a second function and a layer of
|
|
112
|
+
indirection, and buys little.
|
|
113
|
+
|
|
114
|
+
## Pass dependencies at call time
|
|
115
|
+
|
|
116
|
+
A factory that needs something from the outside world (a store, a client, a configured host)
|
|
117
|
+
declares it as a third type parameter and receives it as the second argument to `make`:
|
|
118
|
+
|
|
119
|
+
```typescript
|
|
120
|
+
export const storedOrderFactory = new AsyncMappedFactory<
|
|
121
|
+
OrderParts,
|
|
122
|
+
Order,
|
|
123
|
+
{ orders: OrderStore }
|
|
124
|
+
>(
|
|
125
|
+
() => ({ total: faker.number.int({ min: 100, max: 10_000 }) }),
|
|
126
|
+
async (parts, { orders }) => orders.insert({ id: faker.string.uuid(), ...parts }),
|
|
127
|
+
);
|
|
128
|
+
|
|
129
|
+
// In a test, which decides what to hand it.
|
|
130
|
+
const order = await storedOrderFactory.make({ total: 5000 }, { orders });
|
|
131
|
+
```
|
|
132
|
+
|
|
133
|
+
Dependencies are given at call time rather than held by the factory, and are used as they are given,
|
|
134
|
+
never fetched or awaited. That is what keeps a factory shareable. It holds no state of its own,
|
|
135
|
+
reaches for no ambient state, and cannot depend on what another factory did first. A factory built
|
|
136
|
+
this way can be used in every test file in the codebase and still stand on its own.
|
|
137
|
+
|
|
138
|
+
Keep each dependency as narrow as the factory actually needs. An `OrderStore` is a dependency. The
|
|
139
|
+
whole application never is. A wide dependency is how state starts leaking between tests that were
|
|
140
|
+
supposed to be independent.
|
|
141
|
+
|
|
142
|
+
## Call the factory directly
|
|
143
|
+
|
|
144
|
+
```typescript
|
|
145
|
+
// Anti-pattern. This is make(overrides) with extra steps.
|
|
146
|
+
export function makeCustomer(overrides: Partial<Customer> = {}): Customer {
|
|
147
|
+
return { ...customerFactory.make(), ...overrides };
|
|
148
|
+
}
|
|
149
|
+
```
|
|
150
|
+
|
|
151
|
+
`make(overrides)` already is that function, and it does the job better. Its overrides are partial
|
|
152
|
+
all the way down the nested structure, where the spread above replaces a nested object whole.
|
|
153
|
+
|
|
154
|
+
A wrapper that does anything more than pass overrides through is a signal to read and never to
|
|
155
|
+
write. It usually means one of two things:
|
|
156
|
+
|
|
157
|
+
- **The factory is the wrong shape.** The wrapper is computing something the defaults should be
|
|
158
|
+
computing, or deriving one field from another. Move that into a `DynamicFactory` defaults
|
|
159
|
+
function, which receives the overrides, or into a `MappedFactory` map.
|
|
160
|
+
- **The output type is fighting you.** The wrapper is casting, widening or filling in a field the
|
|
161
|
+
type demands but the test ignores. Fix the type, or use `MappedFactory` so the parts and the
|
|
162
|
+
output are allowed to differ.
|
|
163
|
+
|
|
164
|
+
The same goes for a wrapper that exists to pass a dependency. Declare it on the factory instead.
|
|
165
|
+
|
|
166
|
+
## Factories belong beside the type they construct
|
|
167
|
+
|
|
168
|
+
A library that defines an event, a message or a payload shape should export a factory for it.
|
|
169
|
+
Otherwise every consumer hand-rolls the literal, and every copy drifts.
|
|
170
|
+
|
|
171
|
+
The worked example is an AWS Lambda function URL invocation event, payload format 2.0. It is around
|
|
172
|
+
thirty lines, of which two matter to any given test.
|
|
173
|
+
|
|
174
|
+
```json
|
|
175
|
+
{
|
|
176
|
+
"version": "2.0",
|
|
177
|
+
"routeKey": "$default",
|
|
178
|
+
"rawPath": "/upload",
|
|
179
|
+
"rawQueryString": "part=3",
|
|
180
|
+
"headers": { "host": "abc.lambda-url.eu-west-2.on.aws", "user-agent": "..." },
|
|
181
|
+
"queryStringParameters": { "part": "3" },
|
|
182
|
+
"cookies": ["session=abc"],
|
|
183
|
+
"requestContext": {
|
|
184
|
+
"http": { "method": "POST", "path": "/upload", "sourceIp": "127.0.0.1" }
|
|
185
|
+
},
|
|
186
|
+
"body": "...",
|
|
187
|
+
"isBase64Encoded": false
|
|
188
|
+
}
|
|
189
|
+
```
|
|
190
|
+
|
|
191
|
+
Hand-writing that in three test files gives three copies that drift as the shape changes. It also
|
|
192
|
+
carries a real trap. The path is in the event twice, as `rawPath` and as `requestContext.http.path`,
|
|
193
|
+
and the query string is in it twice, as `rawQueryString` and as `queryStringParameters`. A test that
|
|
194
|
+
sets one and not the other passes against a handler reading the field the test set, and fails in
|
|
195
|
+
production against the same handler reading the other one.
|
|
196
|
+
|
|
197
|
+
A `MappedFactory` removes both problems. The parts are what a test cares about, and the map is what
|
|
198
|
+
fills the event in consistently:
|
|
199
|
+
|
|
200
|
+
```typescript
|
|
201
|
+
import { MappedFactory } from "@kensio/part-factory";
|
|
202
|
+
|
|
203
|
+
interface FunctionUrlRequestParts {
|
|
204
|
+
method: string;
|
|
205
|
+
path: string;
|
|
206
|
+
query: Record<string, string>;
|
|
207
|
+
body?: string;
|
|
208
|
+
}
|
|
209
|
+
|
|
210
|
+
export const functionUrlEventFactory = new MappedFactory<
|
|
211
|
+
FunctionUrlRequestParts,
|
|
212
|
+
FunctionUrlEvent
|
|
213
|
+
>(
|
|
214
|
+
() => ({ method: "GET", path: "/", query: {} }),
|
|
215
|
+
(parts) => ({
|
|
216
|
+
version: "2.0",
|
|
217
|
+
routeKey: "$default",
|
|
218
|
+
rawPath: parts.path,
|
|
219
|
+
rawQueryString: new URLSearchParams(parts.query).toString(),
|
|
220
|
+
queryStringParameters: parts.query,
|
|
221
|
+
// ... and the rest, filled in once.
|
|
222
|
+
requestContext: {
|
|
223
|
+
http: { method: parts.method, path: parts.path, sourceIp: "127.0.0.1" },
|
|
224
|
+
},
|
|
225
|
+
body: parts.body,
|
|
226
|
+
isBase64Encoded: false,
|
|
227
|
+
}),
|
|
228
|
+
);
|
|
229
|
+
|
|
230
|
+
// In a test, the two lines that matter are the two lines written.
|
|
231
|
+
const event = functionUrlEventFactory.make({ method: "POST", path: "/upload" });
|
|
232
|
+
```
|
|
233
|
+
|
|
234
|
+
The path can no longer disagree with itself, because there is one place it is set.
|
|
235
|
+
|
|
236
|
+
Export the factory from the package that owns the type, from a test-support entry point so it does
|
|
237
|
+
not ship in the production bundle. Consumers then get a correct event with one call, and a change to
|
|
238
|
+
the shape is made once.
|
|
@@ -0,0 +1,126 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: skill-template
|
|
3
|
+
description: Scaffold a new agent skill, writing the SKILL.md and its frontmatter to the Agent Skills specification, then wrapping it as a plugin in this repo with package.json, plugin.json and a marketplace entry. Use when the user asks to "add a new skill", "create a skill", "write a SKILL.md" or "start a new plugin", and when checking whether an existing skill is portable between agents.
|
|
4
|
+
license: Apache-2.0
|
|
5
|
+
metadata:
|
|
6
|
+
version: "1.13.1"
|
|
7
|
+
---
|
|
8
|
+
|
|
9
|
+
# Skill template
|
|
10
|
+
|
|
11
|
+
A skill is a directory with a `SKILL.md` in it. That directory is the artefact, and every agent
|
|
12
|
+
reads it. The plugin folder around it in this repository is packaging for one of them.
|
|
13
|
+
|
|
14
|
+
Write the skill first, following the [specification](https://agentskills.io/specification). Then
|
|
15
|
+
wrap it.
|
|
16
|
+
|
|
17
|
+
## The skill directory
|
|
18
|
+
|
|
19
|
+
```
|
|
20
|
+
<skill-name>/
|
|
21
|
+
├── SKILL.md
|
|
22
|
+
├── references/ # optional, loaded only when linked to
|
|
23
|
+
├── scripts/ # optional, anything the skill runs
|
|
24
|
+
└── assets/ # optional, templates and data files
|
|
25
|
+
```
|
|
26
|
+
|
|
27
|
+
Those three subdirectory names come from the specification. An agent that supports skills at all
|
|
28
|
+
knows this shape, whether it reads the directory from `.agents/skills/`, `.claude/skills/`,
|
|
29
|
+
`.github/skills/` or a plugin.
|
|
30
|
+
|
|
31
|
+
**Every path a skill mentions is relative to its own directory.** The directory gets copied out on
|
|
32
|
+
its own, unzipped somewhere unrelated, and installed under a name this repository never sees. A
|
|
33
|
+
command written as `node skills/<skill-name>/scripts/check.mjs` works here and nowhere else. Write
|
|
34
|
+
`node scripts/check.mjs`. `pnpm validate:skills` fails the build on any path that reaches outside
|
|
35
|
+
the skill, which is the check that caught this after it had already shipped once.
|
|
36
|
+
|
|
37
|
+
## Frontmatter
|
|
38
|
+
|
|
39
|
+
```markdown
|
|
40
|
+
---
|
|
41
|
+
name: <skill-name>
|
|
42
|
+
description: <what it does, then when to use it. Include the words and phrases a user would actually type>
|
|
43
|
+
license: Apache-2.0
|
|
44
|
+
metadata:
|
|
45
|
+
version: "0.0.0"
|
|
46
|
+
---
|
|
47
|
+
```
|
|
48
|
+
|
|
49
|
+
- `name` is lowercase, hyphenated, at most 64 characters, and matches the containing directory.
|
|
50
|
+
- `description` is the _only_ thing an agent sees when deciding whether to load the skill. It
|
|
51
|
+
carries the whole triggering burden. State what the skill does, then when to use it, in third
|
|
52
|
+
person. Concrete trigger phrases beat abstract summaries. 1024 characters is the ceiling.
|
|
53
|
+
- `license` and `metadata` travel with the directory. A copy in someone's `.agents/skills/` has no
|
|
54
|
+
package.json beside it, and these are then the only record of what it is and where it came from.
|
|
55
|
+
The release sets `metadata.version`. Never edit that number by hand.
|
|
56
|
+
- The specification allows two more keys. `compatibility` states an environment requirement, such as
|
|
57
|
+
a binary the scripts need. `allowed-tools` restricts the tools the skill may use, and support for
|
|
58
|
+
it varies between agents. Any other key fails validation.
|
|
59
|
+
|
|
60
|
+
Keep `SKILL.md` under 500 lines. It is instructions for an agent, and documentation for a human
|
|
61
|
+
belongs in the README. Push detail into `references/` and link to it. The body stays cheap to load
|
|
62
|
+
and the details are read only when needed.
|
|
63
|
+
|
|
64
|
+
## Wrapping it as a plugin
|
|
65
|
+
|
|
66
|
+
Claude Code installs skills as plugins, so each one here has a plugin folder around it:
|
|
67
|
+
|
|
68
|
+
```
|
|
69
|
+
plugins/<skill-name>/
|
|
70
|
+
├── package.json # npm package: @kensio/<skill-name>
|
|
71
|
+
├── README.md # for humans arriving from npm or GitHub
|
|
72
|
+
├── .claude-plugin/
|
|
73
|
+
│ └── plugin.json # name, version, description, author
|
|
74
|
+
└── skills/
|
|
75
|
+
└── <skill-name>/ # the skill directory above
|
|
76
|
+
```
|
|
77
|
+
|
|
78
|
+
1. Create `plugins/<skill-name>/` following that layout.
|
|
79
|
+
2. Copy `package.json` from an existing plugin, then set `name` to `@kensio/<skill-name>` and
|
|
80
|
+
`repository.directory` to `plugins/<skill-name>`.
|
|
81
|
+
3. Copy `.claude-plugin/plugin.json`, then set `name` and `description`.
|
|
82
|
+
4. Set the `version` in both files, and in the `SKILL.md` frontmatter, to whatever the other plugins
|
|
83
|
+
currently carry. Versions move in lockstep across the whole repo and the release workflow is what
|
|
84
|
+
changes them. Never pick a new number by hand.
|
|
85
|
+
5. Write `skills/<skill-name>/SKILL.md`.
|
|
86
|
+
6. Add an entry to `.claude-plugin/marketplace.json` with a matching `name`, a `source` of
|
|
87
|
+
`"./plugins/<skill-name>"`, and a description.
|
|
88
|
+
7. Run `pnpm check`.
|
|
89
|
+
|
|
90
|
+
Anything under `skills/` ships, because that is what `package.json` lists in `files`. A script the
|
|
91
|
+
skill runs belongs there too, and never at the plugin root.
|
|
92
|
+
|
|
93
|
+
**A plugin folder must be self-contained.** Never reference files outside it with `../`. Plugins are
|
|
94
|
+
copied, zipped and installed standalone, and those paths will not resolve.
|
|
95
|
+
|
|
96
|
+
Nothing else needs telling about the new folder. `scripts/set-version.mjs`,
|
|
97
|
+
`scripts/publish-npm.mjs` and `scripts/build-zips.mjs` all read the `plugins/` directory, so a new
|
|
98
|
+
skill is versioned, bundled into `@kensio/skills`, zipped onto the release and published without
|
|
99
|
+
being listed anywhere else.
|
|
100
|
+
|
|
101
|
+
**A brand new package still needs one manual first publish.** npm trusted publishing cannot create a
|
|
102
|
+
package that does not exist, because the trusted publisher is configured against a package already
|
|
103
|
+
on the registry. The release reports the commands and carries on. See "npm publishing" in the
|
|
104
|
+
repository README.
|
|
105
|
+
|
|
106
|
+
## Prose
|
|
107
|
+
|
|
108
|
+
`pnpm check` runs `pnpm prose`, which fails the build on em dashes, semicolons, and five sentence
|
|
109
|
+
shapes measured against Django, Effective Go, the Rust Book and the Python docs. A new `SKILL.md`
|
|
110
|
+
and `README.md` have to pass it.
|
|
111
|
+
|
|
112
|
+
Load the `technical-prose-style` skill before writing either one. It carries the rules, the
|
|
113
|
+
before-and-after examples, and the evidence for each. Checking a single file while drafting:
|
|
114
|
+
|
|
115
|
+
```bash
|
|
116
|
+
node plugins/technical-prose-style/skills/technical-prose-style/scripts/prose-check.mjs plugins/<skill-name>
|
|
117
|
+
```
|
|
118
|
+
|
|
119
|
+
## Releasing
|
|
120
|
+
|
|
121
|
+
Releasing is automatic. Merging to `main` releases, and the version comes from the pull request
|
|
122
|
+
title. `fix:` for a patch, `feat:` for a minor, `feat!:` or a `BREAKING CHANGE` footer for a major.
|
|
123
|
+
A `docs:` or `chore:` title releases nothing.
|
|
124
|
+
|
|
125
|
+
Every plugin is set to the new version together. A released version means the same commit wherever
|
|
126
|
+
it was installed from.
|