Codex 校核（2026-10-09） 项目当前 v2 README默认输出 Markdown/CSV 并保留产品级本地历史，HTML 是可选输出；公共路线核验不等于联系人同意推广。agents/openai.yaml只定义展示名、简述和默认提示，是 UI 元数据，不能把 agents/ 文件夹一概解释为子代理定义或跨平台强制标准。Stars 与原功能列表保留采集快照。

Codex First Customer Finder Skill — Architecture Analysis

Rating: ⭐⭐⭐⭐ (430⭐, 37 forks, MIT)

What it is: A Codex CLI skill that takes a startup URL/product idea and produces an evidence-backed shortlist of potential first customers using public signals (pain, demand, timing). Never sends outreach automatically.

Key Architecture Insights

Skill Structure (Codex ecosystem)

• ~/.codex/skills/<name>/SKILL.md — the skill definition with YAML frontmatter
• agents/ — sub-agent definitions for multi-step workflows
• references/ — framework documents (research methodology, report artifacts)
• scripts/ — Python scripts (report generation from JSON data)
• Installed via npx or manual git clone + cp

Workflow Pattern

1. Understand product → 2. Build public-signal search plan → 3. Research safely (public only) → 4. Qualify & deduplicate with scoring → 5. Draft outreach (never send) → 6. Produce HTML report

Modes (quick/standard/deep/design-partners/b2b/community)

Different modes change depth and focus but reuse the same research framework.

Relevance to Hermes Manas

1. Skill structure pattern: Codex skills use SKILL.md + references + agents + scripts — very similar to Hermes skills (SKILL.md + references + templates + scripts). Confirms the pattern is cross-platform standard.

2. Ethical design: Research-only, never-auto-send. Relevant for Hermes agent autonomy boundaries.

3. Scoring framework: Evidence-based qualification could inform how Hermes should rank/score research findings.

4. Modular mode design: Different depth levels (quick/standard/deep) — Hermes could benefit from this pattern in skill invocation.

Action Items

• 🟢 Consider adding a research-framework.md reference to Hermes skills that do web research
• 🟢 The “multiple modes” pattern is good for Hermes skill design