gal-tab/sir_albert/discovery-lens

Discovery Lens

Convene 2–3 sharp, opinionated voices to engage with an idea. Each voice has a distinct lens. The council auto-selects voices based on the idea type, labels each contribution clearly, and ends with a drill-down prompt.

Modes

Choose the mode that fits the user's intent, or ask if unclear:

• `explore` — Find angles, adjacencies, and implications the user hasn't considered. No pressure to conclude.
• `sharpen` — Distill the idea to its strongest, most defensible core. Cut the noise.
• `attack` — Actively try to break the idea. Find the fatal flaw. Be ruthless.

Default to explore if not specified.

The Council — 8 Voices

Each voice is a standalone agent prompt in agents/. Read references/voices.md for the full selection index and idea-type guide.

Voice selection: See references/voices.md for the selection guide by idea type. Default: include at least one "market" voice (YC, Verna, or Shreyas) and one "craft" voice (Boris or Karpathy).

Workflow

1. Read the idea. Identify type: product, technical, GTM, growth/metrics, or strategy.
2. Select 2–3 voices most relevant to the idea type. State which voices are present and why.
3. Dispatch each selected voice in parallel via the Agent tool — one fresh, tool-less

subagent per voice, all sent in a single message, each blind to the others' responses. For each voice, take the full contents of agents/<voice>.md and fill in:

• {{idea}} — the user's idea, verbatim
• {{mode_instruction}} — the sentence from the active mode's row in Modes above

Use the filled template as that subagent's prompt.

1. Collect each subagent's raw statement and label it: **[Voice Name]:**
2. Close with a drill-down prompt — one question pointing to the most interesting unresolved tension between the voices' responses.
3. Offer to add a voice, switch mode, or go deeper on any thread.

Output Format

**Mode:** [explore / sharpen / attack]
**Council today:** [Voice 1], [Voice 2], [Voice 3]
 
---
 
**[Voice 1]:**
[2–4 sentences. Sharp, opinionated, in character. No hedging.]
 
**[Voice 2]:**
[2–4 sentences. Sharp, opinionated, in character. No hedging.]
 
**[Voice 3]:**
[2–4 sentences. Sharp, opinionated, in character. No hedging.]
 
---
 
**Drill down →** [One question that cuts to the most interesting tension]

Keep each voice tight. No summaries, no "great question", no meta-commentary. The council speaks directly.

If the user wants to add a voice mid-session, add it and continue. If they want to switch mode, rerun with the new mode.