📦 SaaS Idea Scout — SaaS 思想探索员

SaaS Idea Scout What This Is

A chat-driven 流水线 for discovering and stress-测试 SaaS product ideas. You talk through your domAIn and constrAInts, the 代理 生成s 8 idea 种子s, then fans out 24 sub-代理s across 3 phases — discovery, critique, and evaluation — to produce scored, ranked PRDs with adversarial 验证.

Key features:

Adversarial critique gauntlet — Every idea is championed by a re搜索 代理, attacked by a critic, and objectively scored by an evaluator, all independently.

Parallel 代理ic swarm — 24 sub-代理s 运行 concurrently in batches of 4, completing a full 流水线 in ~30-45 minutes of wall time.

Self-healing — A cron watchdog 监控s 代理 健康 and a retry escalation l添加er 恢复s from 失败s automatically.

Separation of concerns — Discovery, critique, and evaluation are handled by independent 代理s with distinct 角色s and instructions. No 角色 confusion.

上下文ual founder ranking — Phase 5 应用lies founder-性能分析 awareness to rank ideas by what matters for YOUR specific situation.

10-dimension scoring rubric — Structured, transparent evaluation with dimensional scores and narrative synthesis.

5 phases: Ideation → Discovery → Critique → Evaluation → Judgement

8 ideas → 24 sub-代理s → 上下文ually ranked top 3

~30-45 minutes wall time

No 代理 config. No OpenClaw.json changes. 安装s and 运行s through your chat 会话.

When To Use "Find SaaS opportunities in [domAIn]" "Quick 验证 of business ideas in [industry]" "Surface promising product directions before deep re搜索" "I have a problem space — what products could solve it?" "Stress-test these 启动up ideas" Prerequisites

None. 安装 and 运行. Works with any OpenClaw 设置up using the default 模型. For stronger 结果s, use a capable 模型 — a stronger 模型 with higher thinking produces better re搜索 and deeper critiques.

How It Works Chat 代理 (编排器 + your conversation partner) │ ├─ PHASE 1: IDEATION (conversational, ~5 min) │ └─ 4-5 questions → 8 idea 种子s → you confirm │ └─ 创建s .scoutrc on first 运行 (模型/thinking preference) │ ├─ PHASE 2: DISCOVERY (8 sub-代理s, 2 batches of 4, ~10 min) │ ├─ Each 代理 re搜索es one idea, writes a PRD │ └─ Verification: file size + section header 检查s │ ├─ PHASE 3: CRITIQUE (8 sub-代理s, 2 batches of 4, ~8 min) │ ├─ Each 代理 reads one PRD, conducts adversarial re搜索, writes critique │ └─ Verification: section header 检查s → 编排器 merges into dossiers │ ├─ PHASE 4: EVALUATION (8 sub-代理s, 2 batches of 4, ~8 min) │ ├─ Each 代理 reads PRD + critique, writes ho列出ic evaluation with scores │ └─ Verification: section headers + integrity 检查s → 编排器 merges │ └─ PHASE 5: JUDGEMENT (in-会话, ~5 min) ├─ Read all 8 assembled dossiers (PRD + critique + evaluation per idea) └─ 上下文ual ho列出ic ranking → top 3 with narrative justification

A cron watchdog spans all spawn phases (Discovery through Evaluation). It is 创建d after Phase 1 and 移除d after Phase 4 completes. On every wake (completion event or watchdog), the 编排器 运行s its verification procedure from its own 上下文 — it knows which phase and batch it's in.

流水线 Operations Batch-验证 Pattern

All spawn phases (2-4) follow the same pattern:

发送 进度 message for the batch Spawn up to 4 sub-代理s simultaneously (max 4 concurrent, no overlap) 会话s_yield() to wAIt for completion or watchdog wake On every wake, 验证 expected 输出 files exist using exec commands If files are missing: identify which 代理s, re-spawn individually, yield agAIn Repeat until all files for the current batch exist 发送 batch-complete 进度, proceed to next batch

Never proceed to the next batch until ALL files from the current batch are verified.

Yield and Watchdog Interaction

When the 编排器 yields, it wakes for either:

Sub-代理 completion 事件 — 代理s finished their work Cron watchdog fires — the 7-minute interval elapsed

On EVERY wake (regardless of trigger), unconditionally 验证 输出 files. Never assume completion 事件 are authoritative — they can be delayed or dropped.

If all expected files for the current batch exist → proceed. If files are missing but 代理s are still 运行ning → yield agAIn. If files are missing and 代理s are done/stuck → intervene (see re-spawn escalation below).

Verification Commands

Verification uses section-header grep and file size thresholds, following the pattern established in saas-idea-discovery/instructions/验证_phase.sh.

After Phase 2 (Discovery) — 验证 all 8 PRDs:

for f in <运行_dir>/pool/0[1-8]_prd.md; do sz=$(stat -c%s "$f" 2>/dev/null || stat -f%z "$f") [ "$sz" -gt 2048 ] || { echo "FAIL: $(basename $f) too small ($sz bytes)"; continue; } grep -q "^### Problem" "$f" && grep -q "^### Rough Market Size" "$f" \ && grep -q "^### Initial Score" "$f" \ && echo "OK: $(basename $f)" \