Aweskill Creator
The creative loop is yours; the mechanics belong to the aweskill CLI. You interview, draft, test, and revise. The CLI guarantees valid scaffolding, frontmatter, and lifecycle. Read references/writing-guide.md before drafting skill prose.
The Loop
- Capture intent. If the current conversation already shows a workflow worth capturing, extract it from history first (tools used, step order, corrections, input/output formats). Confirm gaps with the user before drafting: what should the skill enable, when should it trigger, what output format, which examples pin the behavior down.
- Check for existing skills first. Run
aweskill find <query> and aweskill store list --verbose. Reuse, fork, or improve an existing skill before writing a new one.
- Scaffold. Run
aweskill store create <name> --description "<trigger description>". The command validates the name, checks for conflicts, and writes a valid SKILL.md skeleton with references/.
- Draft the body. Fill the skeleton following
references/writing-guide.md. Keep the body under ~500 lines; move detail into references/ files and link them from the body.
- Project and test. Run
aweskill agent add skill <name> --global --agent <agent-id>, then try 2–3 realistic test prompts in fresh turns. Projections are symlinks: edits to the central-store copy go live immediately, so you can iterate without re-projecting.
- Validate the mechanics. Run
aweskill doctor fix-skills --skill <name> (dry-run). Fix anything actionable before finishing.
- Improve and repeat. Generalize feedback instead of layering rules, cut guidance that is not pulling its weight, rerun the test prompts. Stop when changes stop helping.
- Finish. Group it with
aweskill bundle add <bundle> <name>, or publish it: copy the skill directory into its own git repo so others can aweskill install owner/repo.
Where the Skill Lives
| Skill type |
Create with |
| Personal, used across agents |
aweskill store create <name> (central store) |
| Repo-specific, shared through git |
aweskill store create <name> --dir <repo>/.agents/skills |
Repo-specific skills stay in the repo; bring one into a machine's central store later with aweskill store install <path>.
Editing an Existing Managed Skill
Installed projections are symlinks to ~/.aweskill/skills/<name>/. Edit the central-store copy directly — every agent projection updates instantly. aweskill update protects those local edits unless --override is passed. Keep the original name and directory name; do not fork a skill into <name>-v2 when improving it.
Escalation
Hand off to $aweskill-doctor when projections break or duplicates appear after repeated create/project cycles, and to $aweskill for routine install, bundle, and projection work outside authoring.
References
references/writing-guide.md — trigger descriptions, progressive disclosure, writing style, test prompts, and the improve loop
1---2name: aweskill-creator3description: Use when creating, authoring, or improving a skill: scaffold a new skill, draft SKILL.md content, write trigger descriptions, test with realistic prompts, iterate wording, turn a repeated workflow from the current conversation into a reusable skill, or customize an installed skill. 中文触发词：制作技能、创建技能、新建skill、写一个技能、把这段流程做成技能、技能创作、改进技能、改进触发词、定制已安装技能。4---56# Aweskill Creator78The creative loop is yours; the mechanics belong to the aweskill CLI. You interview, draft, test, and revise. The CLI guarantees valid scaffolding, frontmatter, and lifecycle. Read `references/writing-guide.md` before drafting skill prose.910## The Loop11121. **Capture intent.** If the current conversation already shows a workflow worth capturing, extract it from history first (tools used, step order, corrections, input/output formats). Confirm gaps with the user before drafting: what should the skill enable, when should it trigger, what output format, which examples pin the behavior down.132. **Check for existing skills first.** Run `aweskill find <query>` and `aweskill store list --verbose`. Reuse, fork, or improve an existing skill before writing a new one.143. **Scaffold.** Run `aweskill store create <name> --description "<trigger description>"`. The command validates the name, checks for conflicts, and writes a valid SKILL.md skeleton with `references/`.154. **Draft the body.** Fill the skeleton following `references/writing-guide.md`. Keep the body under ~500 lines; move detail into `references/` files and link them from the body.165. **Project and test.** Run `aweskill agent add skill <name> --global --agent <agent-id>`, then try 2–3 realistic test prompts in fresh turns. Projections are symlinks: edits to the central-store copy go live immediately, so you can iterate without re-projecting.176. **Validate the mechanics.** Run `aweskill doctor fix-skills --skill <name>` (dry-run). Fix anything actionable before finishing.187. **Improve and repeat.** Generalize feedback instead of layering rules, cut guidance that is not pulling its weight, rerun the test prompts. Stop when changes stop helping.198. **Finish.** Group it with `aweskill bundle add <bundle> <name>`, or publish it: copy the skill directory into its own git repo so others can `aweskill install owner/repo`.2021## Where the Skill Lives2223| Skill type | Create with |24|---|---|25| Personal, used across agents | `aweskill store create <name>` (central store) |26| Repo-specific, shared through git | `aweskill store create <name> --dir <repo>/.agents/skills` |2728Repo-specific skills stay in the repo; bring one into a machine's central store later with `aweskill store install <path>`.2930## Editing an Existing Managed Skill3132Installed projections are symlinks to `~/.aweskill/skills/<name>/`. Edit the central-store copy directly — every agent projection updates instantly. `aweskill update` protects those local edits unless `--override` is passed. Keep the original `name` and directory name; do not fork a skill into `<name>-v2` when improving it.3334## Escalation3536Hand off to `$aweskill-doctor` when projections break or duplicates appear after repeated create/project cycles, and to `$aweskill` for routine install, bundle, and projection work outside authoring.3738## References3940- `references/writing-guide.md` — trigger descriptions, progressive disclosure, writing style, test prompts, and the improve loop
Webioinfo01