Skip to content
Back to skills

Discovery

ASecurity

Interview-first discovery routine — run an interactive spec interview BEFORE building anything non-trivial. Use when you type /discovery, or at the start of any larger task (new site/app/feature/ dashboard/report/tool, a multi-step build, or a vague "make it better" ask). Turns a fuzzy ask into a confirmed spec + plan before code. Skip for one-line lookups, trivial fixes, or when you already answered this turn.

  • 2 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added September 19, 2026
ai-agentsapi

Works with

  • api

Security analysis

A100/100

Scanned September 19, 2026

npx -y skills add SirCharan/second-brain --skill discovery --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Discovery?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Discovery
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/sircharan-discovery/badge)](https://www.skillsdirectory.com/skills/sircharan-discovery)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: discovery
description: >
  Interview-first discovery routine — run an interactive spec interview BEFORE building anything
  non-trivial. Use when you type /discovery, or at the start of any larger task (new site/app/feature/
  dashboard/report/tool, a multi-step build, or a vague "make it better" ask). Turns a fuzzy ask into a confirmed spec + plan
  before code. Skip for one-line lookups, trivial fixes, or when you already answered this turn.
---

# /discovery — interview-first spec, then plan

the standing rule (CLAUDE.md + working-with-claude): get a full download before building. This
operationalizes it. Adapted from Anthropic's "interview-first spec" guidance.

## Steps
1. **Read the room first.** Skim the relevant code/memory (spawn an Explore subagent if scope is
   uncertain) so questions are informed, not generic. Recall: `/second-brain pull <terms>` if stuck.
2. **Interview with `AskUserQuestion`** — 2–4 questions, `multiSelect: true` where apt. Cover the five
   axes, and **recommend a default in each** (first option, "(Recommended)") so it's a confirmation,
   not a blank:
   - **Scope** — what's in / explicitly out.
   - **Target environment** — repo, stack, deploy target, live vs sandbox (esp. money systems).
   - **Output shape** — files, UI, report, API; format + where it lands.
   - **Tradeoffs** — speed vs thoroughness, cost ceiling (OpenRouter cap), token budget.
   - **Done-criteria** — how you will know it's finished + how to verify.
   Surface the ambiguities you might not have spelled out — that's the value.
3. **Restate the spec** in 2–3 lines from the answers; note assumptions taken.
4. **`TaskCreate` a plan** (for 3+ step work) and execute step-by-step, marking tasks inline.
5. **Hand the work a verification check** (test/build/live URL) so it self-checks, not guesses "done".

## Notes
- If it's genuinely small, say so and just do it — don't ceremony a one-liner.
- For large multi-phase builds, escalate to `gtan-workflow`; for the pure UI pipeline, `/design-build`.
- Reach for this yourself at the start of any larger task; nothing forces it on you.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.