Auto Skill Creator

Auto Skill Creator

Turns complex problem-solving into reusable skills — automatically.

When to Trigger

After completing a multi-step task, evaluate:

1. Complexity — Did it take 3+ tool calls or require debugging?
2. Novelty — Was the solution non-obvious or undocumented?
3. Recurrence — Will this pattern likely happen again?

If 2 of 3 are true → create a skill.

Process

1. Extract the Pattern

From the completed task, identify:

• Problem class — what category of problem was this? (e.g., "proxy crash recovery", "WhatsApp reconnect")
• Key steps — the minimal sequence that solves it
• Gotchas — what didn't work or was misleading
• Prerequisites — what tools/access/knowledge is needed

2. Check for Duplicates

Before creating:

grep -rl "" skills/*/SKILL.md

If a similar skill exists, update it instead of creating a new one.

3. Create the Skill

Structure:

skills//
├── SKILL.md          # Frontmatter + instructions
├── scripts/          # Only if deterministic code was part of the solution
└── references/       # Only if reference material is needed

Rules:

• name: lowercase, hyphenated, verb-led (e.g., debug-litellm-proxy, recover-whatsapp-gateway)
• description: max 1024 chars, includes what + when to use
• Body: / skills/_manifest.json

git commit -m "auto-skill: — " git push

### 6. Log It

```bash
python3 tools/eval_tracker.py log proactive_action "auto-created skill: " 2

And add to daily notes:

[HH:MM IL] Auto-created skill  from .

Quality Gate

Before committing, verify:

• [ ] Frontmatter has name and description
• [ ] Description explains both WHAT and WHEN
• [ ] Body is "Did I just solve something hard? Should this be a skill?"

If yes, trigger auto-skill-creator before closing the task.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: netanel-abergel
• Source: netanel-abergel/pa-skills
• License: MIT

Install and usage instructions live in the source repository linked above.