create-skill

Name: create-skill
Author: Bennyoooo

21stars2forks

How to install create-skill?

bennyoooo/skillmaxxing/create-skill

$npx -y skills add Bennyoooo/skillmaxxing --skill create-skill

Installs into the current project.

Summary

Create a new reusable agent skill — either explicitly ("turn this workflow into a skill") or by reflecting on a just-completed task that you might do again. Synthesizes SKILL.md, scripts, and a real eval scaffold, then stages it for review before committing. Use when the user ask

Files · 1

View on GitHub

SKILL.md

1	# create-skill
2
3	Crystallize a workflow into a durable, tested skill. The CLI is model-agnostic: you synthesize the content (name, body, scripts, real eval tasks); the CLI stages, smoke-tests, and commits it atomically.
4
5	## Two entry points
6
7	1. Explicit — the user says "turn X into a skill."
8	2. Reflection — after completing a non-trivial task, consider whether it is reusable. Fire this sparingly (review fatigue is real): only when the work was non-trivial (several steps, a fixed bug, a discovered workflow) AND plausibly recurs. When in doubt, don't interrupt.
9
10	## Step 0 — Prefer update over create
11
12	Before creating, search for an existing skill that already covers this:
13
14	```bash
15	scripts/discover.sh "<capability>" --json
16	```
17
18	If a close match exists, update or optimize it instead of making a near-duplicate. The skillify step also runs this check and will refuse with a suggestion unless you pass `--new`.
19
20	## Step 1 — Synthesize a draft
21
22	Write a draft JSON file. The eval scaffold MUST contain real, scorable tasks (not a stub) — pick a scorer per task: `exact`/`normalized`/`code-exec`/`success-signal` for deterministic outputs, `agent-judge` (with a `rubric`) for prose/judgment skills.
23
24	```json
25	{
26	"name": "release-notes",
27	"description": "Draft release notes from a git log range.",
28	"body": "# release-notes\n\n...instructions...\n",
29	"tools": ["Bash"],
30	"scripts": [{ "path": "scripts/changelog.sh", "content": "#!/usr/bin/env bash\n..." }],
31	"eval": {
32	"skill": "release-notes",
33	"tasks": [
34	{ "id": "happy", "input": "v1.0..v1.1", "scorer": "agent-judge", "rubric": "Groups changes by type; no raw SHAs; user-facing tone." }
35	]
36	},
37	"smokeTest": ["bash", "scripts/changelog.sh", "--help"]
38	}
39	```
40
41	## Step 2 — Stage (with the human approval gate)
42
43	```bash
44	scripts/skillify.sh --draft draft.json # stage; smoke test skipped unless authorized
45	scripts/skillify.sh --draft draft.json --allow-exec # stage AND run the smoke test in the sandbox
46	```
47
48	Only pass `--allow-exec` after the user has reviewed the generated scripts — a freshly synthesized skill is `trusted: false`, and running its code is a deliberate, user-authorized step. Show the staged `SKILL.md` and scripts to the user and get explicit approval before committing.
49
50	## Step 3 — Commit
51
52	```bash
53	scripts/skillify.sh --commit <name> [--global] [--agent <name>]
54	```
55
56	This installs the skill (`trusted: false`) and clears the draft. Staged drafts persist across sessions — resume with `--list-drafts` then `--commit`.
57
58	## Safety
59
60	- The smoke-test/exec gate is yours to honor: never pass `--allow-exec` without fresh user authorization for the specific scripts.
61	- Do not commit a skill whose smoke test failed; fix the draft and re-stage.

People who liked this also viewed

6 picks

Type

create-skill

21stars2forks

View on GitHub

How to install create-skill?

bennyoooo/skillmaxxing/create-skill

$npx -y skills add Bennyoooo/skillmaxxing --skill create-skill

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# create-skill
2
3	Crystallize a workflow into a durable, tested skill. The CLI is model-agnostic: you synthesize the content (name, body, scripts, real eval tasks); the CLI stages, smoke-tests, and commits it atomically.
4
5	## Two entry points
6
7	1. Explicit — the user says "turn X into a skill."
8	2. Reflection — after completing a non-trivial task, consider whether it is reusable. Fire this sparingly (review fatigue is real): only when the work was non-trivial (several steps, a fixed bug, a discovered workflow) AND plausibly recurs. When in doubt, don't interrupt.
9
10	## Step 0 — Prefer update over create
11
12	Before creating, search for an existing skill that already covers this:
13
14	```bash
15	scripts/discover.sh "<capability>" --json
16	```
17
18	If a close match exists, update or optimize it instead of making a near-duplicate. The skillify step also runs this check and will refuse with a suggestion unless you pass `--new`.
19
20	## Step 1 — Synthesize a draft
21
22	Write a draft JSON file. The eval scaffold MUST contain real, scorable tasks (not a stub) — pick a scorer per task: `exact`/`normalized`/`code-exec`/`success-signal` for deterministic outputs, `agent-judge` (with a `rubric`) for prose/judgment skills.
23
24	```json
25	{
26	"name": "release-notes",
27	"description": "Draft release notes from a git log range.",
28	"body": "# release-notes\n\n...instructions...\n",
29	"tools": ["Bash"],
30	"scripts": [{ "path": "scripts/changelog.sh", "content": "#!/usr/bin/env bash\n..." }],
31	"eval": {
32	"skill": "release-notes",
33	"tasks": [
34	{ "id": "happy", "input": "v1.0..v1.1", "scorer": "agent-judge", "rubric": "Groups changes by type; no raw SHAs; user-facing tone." }
35	]
36	},
37	"smokeTest": ["bash", "scripts/changelog.sh", "--help"]
38	}
39	```
40
41	## Step 2 — Stage (with the human approval gate)
42
43	```bash
44	scripts/skillify.sh --draft draft.json # stage; smoke test skipped unless authorized
45	scripts/skillify.sh --draft draft.json --allow-exec # stage AND run the smoke test in the sandbox
46	```
47
48	Only pass `--allow-exec` after the user has reviewed the generated scripts — a freshly synthesized skill is `trusted: false`, and running its code is a deliberate, user-authorized step. Show the staged `SKILL.md` and scripts to the user and get explicit approval before committing.
49
50	## Step 3 — Commit
51
52	```bash
53	scripts/skillify.sh --commit <name> [--global] [--agent <name>]
54	```
55
56	This installs the skill (`trusted: false`) and clears the draft. Staged drafts persist across sessions — resume with `--list-drafts` then `--commit`.
57
58	## Safety
59
60	- The smoke-test/exec gate is yours to honor: never pass `--allow-exec` without fresh user authorization for the specific scripts.
61	- Do not commit a skill whose smoke test failed; fix the draft and re-stage.

People who liked this also viewed

6 picks

Type