Wehuman01/aweskill

# Aweskill

Use `aweskill` CLI directly. Do not add wrapper scripts unless the CLI is missing a needed capability.

## Intent Router

Match the user's intent to a task domain, then follow the workflow below.

| User intent | Domain | First command |
|---|---|---|
| "Find a skill for X", "search skills", "install from GitHub" | Source Lifecycle | `aweskill find <query>` |
| "Scan and import local skills", "what's in the store", "remove a skill from the store" | Store Work | `aweskill store list --verbose` |
| "Create a bundle", "add skills to a bundle", "show bundle contents" | Bundle Work | `aweskill bundle list --verbose` |
| "Give Codex skill X", "project a bundle to Cursor", "remove a projection" | Projection Work | `aweskill agent list --verbose` |

# 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

1. **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.
2. **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.
3. **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/`.
4. **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.
5. **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.
6. **Validate the mechanics.** Run `aweskill doctor fix-skills --skill <name>` (dry-run). Fix anything actionable before finishing.
7. **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.
8. **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`.

# Aweskill Doctor

Diagnose first. Mutate only with `--apply`. Verify after every mutation.

## Symptom Workflows

### Agent Cannot See a Skill

The user says a skill should be available but the agent doesn't show it.

```bash
# 1. Check what's projected
aweskill agent list --global --agent <id> --verbose