discovery-eyes
GitHub模拟诉讼证据开示视角，分析团队沟通记录以进行预防性合规培训。识别法律风险模式，提供谨慎写作的指导原则，旨在提升沟通卫生，严禁用于隐瞒或销毁证据。
Trigger Scenarios
Install
npx skills add mohitagw15856/pm-claude-skills --skill discovery-eyes -g -y
SKILL.md
Frontmatter
{
"name": "discovery-eyes",
"description": "Read your team's messages the way opposing counsel would in litigation discovery — prevention training that makes communication hygiene visceral. Use when asked how would our Slack look in discovery, train my team on communication hygiene, review this thread like a plaintiff's lawyer, or what shouldn't we put in writing. Produces the highlighted-exhibit reading of sample messages, the patterns that create legal risk, and a debrief with the write-it-this-way rules — strictly for prevention, never for concealment."
}
Discovery Eyes Skill
Every message your team writes is a potential exhibit with a highlighter across it, read aloud years later, stripped of tone and context. This skill performs that reading on sample messages — jokes that become admissions, speculation that becomes knowledge, "delete this after reading" that becomes the whole case — as prevention training. Its purpose is writing carefully and acting properly, never hiding, deleting, or evading: concealment is both the crime and the cover-up, and this skill refuses to help with either.
What This Skill Produces
- The exhibit reading — supplied sample messages, highlighted and characterized as a plaintiff's lawyer would
- The pattern list — the recurring habits that manufacture legal risk out of ordinary work
- The debrief — write-it-this-way rules that keep candor AND hygiene, plus when-to-pick-up-the-phone guidance
Required Inputs
Ask for these if not provided:
- Sample messages/threads — real (sanitized) or representative of the team's style
- The context — industry and the risk surfaces that matter (employment, IP, competition, safety, securities)
- The audience — engineers, sales, execs — the patterns differ by tribe
Framework: How the Highlighter Reads
- Tone strips off: sarcasm, hyperbole, and dark jokes read literally ("this feature will literally kill someone 😂" reads exactly as written, without the emoji's protection).
- Speculation becomes knowledge: "I bet the numbers are inflated" reads as awareness-at-the-time. The hygiene rule: state facts and questions, not unverified conclusions.
- The fatal phrases: "delete this," "keep this off email," "per our conversation" (after something sensitive), "I know we're not supposed to, but…" — each is a case-builder regardless of what follows.
- Casual legal conclusions: non-lawyers writing "this is definitely infringement/discriminatory/illegal" create exhibits lawyers then own. Rule: describe behavior, route legal characterization to counsel.
- The honest core: good hygiene is NOT saying less truth — it's precision (facts over vibes), routing (privileged questions to counsel properly), and escalation (real concerns raised through channels that fix them — a concern raised and addressed reads well in discovery; a concern joked about and dropped reads terribly).
Output Format
Exhibit Reading: [team/context]
Prevention training — a plausible adversarial reading of sample messages. This skill does not assist with deleting, concealing, or evading preservation obligations; if litigation is reasonably anticipated, preservation duties apply — see counsel.
The Exhibits
[message, quoted] Highlighted as: [how it reads in a filing] · The pattern: [which habit produced it]
The Pattern List
[The 4–6 recurring habits in these samples, each with its risk mechanism]
Debrief — write it this way
| Instead of | Write | Why |
|---|---|---|
| [Plus: the pick-up-the-phone list (what belongs in synchronous conversation — decisions still get documented properly afterward), and the escalate-properly note: raised-and-fixed is the best exhibit there is] |
Quality Checks
- Every reading traces to a supplied message — no invented exhibits
- The prevention-not-concealment banner appears in the artifact
- Rewrites preserve the truth content — hygiene is precision, not omission
- Legal-conclusion language is routed to counsel, not softened into nothing
- The raised-and-fixed principle appears — the goal is better conduct, not quieter records
Anti-Patterns
- Do not advise deleting, auto-expiring, or moving topics off-channel to evade records — that request ends the exercise and goes to counsel
- Do not train people to stop reporting problems — suppressed concerns are worse in court AND in reality
- Do not sanitize into meaninglessness — a team afraid to write anything ships nothing
- Do not perform legal analysis — this is communication training; law belongs to lawyers
- Do not read real named individuals' messages punitively — samples train teams; this is not a surveillance tool
Version History
- 54fad50 Current 2026-07-19 13:11
