make-skill-en

Name: make-skill-en
Author: agentscope-ai

24kstars2.8kforks

How to install make-skill-en?

agentscope-ai/qwenpaw/make-skill-en

$npx -y skills add agentscope-ai/QwenPaw --skill make-skill-en

Installs into the current project.

Summary

Use this skill when sedimenting a session into a reusable workspace skill. Triggers when the user wants to turn the current conversation, workflow, or troubleshooting path into a SKILL.md. Phrases like 'turn this into a skill', 'remember how I did X', 'save this workflow', 'make

Files · 1

View on GitHub

SKILL.md

1	<!--
2	Inspired by Anthropic's `skill-creator` skill (the "creating a skill"
3	portion in particular). Rewritten for QwenPaw.
4	Credit: https://github.com/anthropics/skills/blob/main/skill-creator/SKILL.md
5	-->
6
7	# Make Skill
8
9	Turn the current session into a reusable workspace skill.
10
11	You orchestrate a two-phase flow:
12
13	* Phase A. Propose a compact plan, yield the turn for user approval.
14	* Phase B. On approval, write the full SKILL.md body based on THIS
15	conversation, then persist via `materialize_skill`.
16
17	Do not call `write_file` to create the SKILL.md or any auxiliary files
18	(scripts, JSON, etc.) directly. All initial file creation must go through
19	`materialize_skill` (via the `body` and `extra_files` parameters), which runs
20	the security scanner and writes the manifest atomically. After successful
21	creation, use `edit_file` to modify existing files if needed.
22
23	## Step 0. Determine the focus and derive a skill name
24
25	### 0a. Determine the focus
26
27	Two invocation paths:
28
29	* `/make-skill <focus>`. The focus follows the command verbatim.
30	* Natural language ("turn this into a skill", "save this workflow",
31	"把刚才的 X 流程变成 skill"). Derive a short focus phrase from the
32	conversation topic the user wants to capture. If ambiguous, ask a
33	one-line clarification first.
34
35	### 0b. Derive the skill name
36
37	Derive the skill name from focus with this exact rule:
38
39	```
40	skill_name = "-".join(focus.split())
41	```
42
43	Internal whitespace (space, tab, full-width space, multiple spaces) collapses
44	to a single `-`. Other characters stay as is.
45
46	Examples:
47
48	* `cooking` → `cooking`
49	* `view image debug` → `view-image-debug`
50	* `烹饪食谱` → `烹饪-食谱`
51	* `Stock Price` → `Stock-Price` (case preserved)
52
53	Use this `skill_name` consistently as `plan.name` in Step 1 and as the
54	`name=` argument to `materialize_skill` in Step 3.
55
56	## Step 1. Propose the plan and yield for approval
57
58	Call `create_plan` with all four required arguments
59	(`name`, `description`, `expected_outcome`, `subtasks`):
60
61	* `name`: the normalised `skill_name` from Step 0.
62	* `description`: a COMPACT preview the user reviews. Two parts:
63	* Part 1: Trigger preview. 2 to 4 sentences, plain language. Cover
64	all three of:
65	* Goal. The end result this skill produces.
66	* Trigger. User phrasings and contexts that should invoke it. Be
67	a bit pushy on synonyms.
68	* I/O. What inputs it expects, what outputs it produces.
69	Not yet SKILL.md frontmatter format; that gets distilled later.
70	* Part 2: Step outline and batch plan. Two parts:
71	* Step outline. Numbered list, one short verb phrase per line.
72	No per-step detail, no parameters, no error handling, no
73	sub-bullets, no `##` sub-headings. Just the shape, so the user
74	can judge ordering and scope. Draw step names from what actually
75	happened in THIS conversation. Don't fabricate; omit anything
76	not grounded in the conversation.
77	Example layout (do NOT copy this content):
78	```
79	1. <verb phrase, ~5-10 words>
80	2. <verb phrase, ~5-10 words>
81	3. <…>
82	```
83	* Batch plan. Briefly describe how the steps above will be
84	organised into `run_tool_batch` JSON files:
85	* Which steps chain into one batch (or need to be split into
86	multiple batch files).
87	* Which intermediate steps would normally require agent
88	involvement but can actually be replaced by scripts (regex
89	matching, keyword filtering, JSON parsing, etc.), reducing
90	agent interaction and keeping more steps in the automated
91	batch.
92	* List the expected files, e.g.:
93	```
94	scripts/main.json — main batch workflow
95	scripts/parse.py — parse snapshot to extract target content
96	```
97	The goal is to replace as much agent-in-the-loop judgement as
98	possible with scripts, so the skill executes with a single
99	`run_tool_batch` call and avoids multi-round agent-tool
100	interaction. This lets the user see the batch structure before
101	approving.
102	* `expected_outcome` (plan-level, REQUIRED — distinct from the
103	subtask's `expected_outcome`): one concrete sentence about what
104	success looks like for the whole skill creation. Use the literal
105	string `"A new workspace skill <skill_name> is created, enabled, and
106	invocable via /<skill_name>."` with `<skill_name>` substituted.
107	* `subtasks`: a list with a single subtask:
108	* `name`: `"Write and materialize skill"`
109	* `description`: `"Write the SKILL.md b

People who liked this also viewed

6 picks

Type

make-skill-en

24kstars2.8kforks

View on GitHub

How to install make-skill-en?

agentscope-ai/qwenpaw/make-skill-en

$npx -y skills add agentscope-ai/QwenPaw --skill make-skill-en

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	<!--
2	Inspired by Anthropic's `skill-creator` skill (the "creating a skill"
3	portion in particular). Rewritten for QwenPaw.
4	Credit: https://github.com/anthropics/skills/blob/main/skill-creator/SKILL.md
5	-->
6
7	# Make Skill
8
9	Turn the current session into a reusable workspace skill.
10
11	You orchestrate a two-phase flow:
12
13	* Phase A. Propose a compact plan, yield the turn for user approval.
14	* Phase B. On approval, write the full SKILL.md body based on THIS
15	conversation, then persist via `materialize_skill`.
16
17	Do not call `write_file` to create the SKILL.md or any auxiliary files
18	(scripts, JSON, etc.) directly. All initial file creation must go through
19	`materialize_skill` (via the `body` and `extra_files` parameters), which runs
20	the security scanner and writes the manifest atomically. After successful
21	creation, use `edit_file` to modify existing files if needed.
22
23	## Step 0. Determine the focus and derive a skill name
24
25	### 0a. Determine the focus
26
27	Two invocation paths:
28
29	* `/make-skill <focus>`. The focus follows the command verbatim.
30	* Natural language ("turn this into a skill", "save this workflow",
31	"把刚才的 X 流程变成 skill"). Derive a short focus phrase from the
32	conversation topic the user wants to capture. If ambiguous, ask a
33	one-line clarification first.
34
35	### 0b. Derive the skill name
36
37	Derive the skill name from focus with this exact rule:
38
39	```
40	skill_name = "-".join(focus.split())
41	```
42
43	Internal whitespace (space, tab, full-width space, multiple spaces) collapses
44	to a single `-`. Other characters stay as is.
45
46	Examples:
47
48	* `cooking` → `cooking`
49	* `view image debug` → `view-image-debug`
50	* `烹饪食谱` → `烹饪-食谱`
51	* `Stock Price` → `Stock-Price` (case preserved)
52
53	Use this `skill_name` consistently as `plan.name` in Step 1 and as the
54	`name=` argument to `materialize_skill` in Step 3.
55
56	## Step 1. Propose the plan and yield for approval
57
58	Call `create_plan` with all four required arguments
59	(`name`, `description`, `expected_outcome`, `subtasks`):
60
61	* `name`: the normalised `skill_name` from Step 0.
62	* `description`: a COMPACT preview the user reviews. Two parts:
63	* Part 1: Trigger preview. 2 to 4 sentences, plain language. Cover
64	all three of:
65	* Goal. The end result this skill produces.
66	* Trigger. User phrasings and contexts that should invoke it. Be
67	a bit pushy on synonyms.
68	* I/O. What inputs it expects, what outputs it produces.
69	Not yet SKILL.md frontmatter format; that gets distilled later.
70	* Part 2: Step outline and batch plan. Two parts:
71	* Step outline. Numbered list, one short verb phrase per line.
72	No per-step detail, no parameters, no error handling, no
73	sub-bullets, no `##` sub-headings. Just the shape, so the user
74	can judge ordering and scope. Draw step names from what actually
75	happened in THIS conversation. Don't fabricate; omit anything
76	not grounded in the conversation.
77	Example layout (do NOT copy this content):
78	```
79	1. <verb phrase, ~5-10 words>
80	2. <verb phrase, ~5-10 words>
81	3. <…>
82	```
83	* Batch plan. Briefly describe how the steps above will be
84	organised into `run_tool_batch` JSON files:
85	* Which steps chain into one batch (or need to be split into
86	multiple batch files).
87	* Which intermediate steps would normally require agent
88	involvement but can actually be replaced by scripts (regex
89	matching, keyword filtering, JSON parsing, etc.), reducing
90	agent interaction and keeping more steps in the automated
91	batch.
92	* List the expected files, e.g.:
93	```
94	scripts/main.json — main batch workflow
95	scripts/parse.py — parse snapshot to extract target content
96	```
97	The goal is to replace as much agent-in-the-loop judgement as
98	possible with scripts, so the skill executes with a single
99	`run_tool_batch` call and avoids multi-round agent-tool
100	interaction. This lets the user see the batch structure before
101	approving.
102	* `expected_outcome` (plan-level, REQUIRED — distinct from the
103	subtask's `expected_outcome`): one concrete sentence about what
104	success looks like for the whole skill creation. Use the literal
105	string `"A new workspace skill <skill_name> is created, enabled, and
106	invocable via /<skill_name>."` with `<skill_name>` substituted.
107	* `subtasks`: a list with a single subtask:
108	* `name`: `"Write and materialize skill"`
109	* `description`: `"Write the SKILL.md b

People who liked this also viewed

6 picks

Type