# Skill Document Self-Correction

When you discover that a skill document is incomplete, incorrect, or vague during task execution, you must update it immediately.

## When to Trigger

- A skill's instructions don't match reality (wrong API endpoint, incorrect flag, missing step)
- A skill references a tool/service that doesn't exist or has changed
- A skill's output format doesn't match what downstream consumers expect
- A skill is missing critical information you had to figure out independently
- A skill's cost/rate-limit info is outdated

## Correction Process

1. **Identify** — Note the specific discrepancy (what the skill says vs. what's actually true)
2. **Fix** — Edit the skill's SKILL.md directly with the correct information
3. **Log** — Add a correction entry at the bottom of the skill file:
   ```
   ## Correction Log
   - [YYYY-MM-DD] [Agent Name]: [What was wrong] → [What was fixed]
   ```
4. **Report** — Notify Alfred via Telegram that a skill was corrected, including:
   - Skill name
   - What was wrong
   - What was fixed
   - How it was discovered

## Rules

- Never silently work around a broken skill instruction — fix the doc
- If unsure whether something is wrong, add a `> ⚠️ UNVERIFIED:` note rather than changing it
- Do not remove information — mark deprecated content with `~~strikethrough~~` and add the correction below
- Keep corrections minimal and factual — no commentary
- If a skill needs major restructuring (not just a fix), flag it to Devin for the weekly skill incubation review

## Examples

**Wrong API endpoint:**
```
## Correction Log
- 2026-02-24 Rick: Crustdata endpoint was /v1/company → corrected to /v2/company/search
```

**Missing step:**
```
## Correction Log
- 2026-02-24 Alfred: Added step to run compute:ready after analyze:gbp (was missing from pipeline)
```