adk-explainlisted
Install: claude install-skill sujeet-pro/agents-devkit
# adk-explain
Advisor. Teach, don't pick.
**Global skill** — mostly transient; if any artifact is produced (a written decision-tree, a referenced comparison table) it goes to `$ADK_DATA_HOME/explain/<ts>/`. Never writes to the cwd repo.
`--depth deep` controls how much the answer explains. `--deep` controls the harness model profile per `shared/model-depth.md`. `--detailed` asks for more evidence gathering before teaching the choice.
## When invoked
- Directly: `/adk-explain "what's the difference between hexagonal and onion architecture?"`
- Hand-off: another skill detected "I don't know" / "you decide" / "not sure" in user response.
- From explicit jargon: another skill encounters a term the user hasn't seen before.
## Workflow
```
Phase 0 — context-gather
- If --from <skill> is set, the calling skill passed the question + options + context
- If standalone, parse the topic from the user's prompt; pull related entries from $ADK_DATA_HOME/improve/learning/
Phase 1 — restate
- Restate the question in plain English; define any jargon inline once
- If the topic is outside adk's scope: refuse with a pointer
Phase 2 — explain options
- 2–4 options with the consequence of each: "if you pick X, then Y"
- One marked recommended IF the user has ≥3 prior matches in their decision log; else "no recommendation"
- Quote 2–3 evidence lines from prior decision-log entries verbatim, with task-slugs
Phase 3 — tie-breaker
- The one question whose answer makes the choi