thinking-discoverlisted
Install: claude install-skill seankoji-com/claude-plugins
# Thinking: Discovery Phase
Step 1 of Rensin's three-step process. Run an interrogation, not a survey. The goal is not to
collect the requirements the user already has — it is to find the ones they haven't thought of.
In Claude Code, `/elephant-goldfish:thinking` does this with deterministic state handling and
GitHub publishing. This skill is the portable path: same process, no scripts required.
## Step 1: Resolve the topic from disk, not from memory
List `thinking/` (e.g. `ls thinking/ 2>/dev/null`). Never rely on conversation history to know
which topics exist.
- If topic folders exist, show them and ask whether this is new or a continuation.
- For a new topic, agree a short kebab-case slug (`buy-family-car`, `q3-vendor-report`). The
working folder is `thinking/<topic-slug>/`.
- If `thinking/<topic-slug>/discovery.md` exists, summarise it briefly and ask whether to
refine or restart. Never silently overwrite.
## Step 2: Settle the output type
Ask whether this ends in a **research document** (a standalone answer to a question) or an
**implementation plan** (a plan that becomes code). This picks which probe bank to read and
how the eventual handoff is framed, so it changes the whole conversation — don't default it.
Record it in `thinking/<topic-slug>/meta.json` as `{"slug": ..., "output_type": ...}` so later
phases resolve it from disk rather than memory.
## Step 3: Announce the mode shift
Before the first question, say how this phase works and why, in plain lan