AgentSkillsCN

me-network-crm

在`05_Resources/network/`下的MEMO仓库中，录入、更新并查询个人网络/联系人数据，包括通过PRISM式的分解方式，从外部信息源（截图/聊天记录/个人资料）中提取信息，结合时间感知的生命阶段推断（“阶段锚点”），以及相对“我”的时间差（年龄/毕业时间）进行数据整理。当用户需要从任意原始情境中添加或更新人物卡片、提取事实/推论/行动、计算相对阶段/年龄差异，或通过CLI关键词搜索匹配人物时，可调用此技能。

SKILL.md
--- frontmatter
name: me-network-crm
description: Record (录入), update, and query your personal network/contact data in a MEMO repo under `05_Resources/network/`, including ingesting external info sources (screenshots/chats/profiles) via PRISM-style decomposition, time-aware life-stage inference (`阶段锚点`), and relative-to-Me deltas (age / graduation). Use when asked to add/update a person card from any raw context, extract facts/inferences/actions, compute relative stage/age differences, or search for matching people via tags/keywords using CLI.

Me Network CRM

Files and conventions (source of truth)

  • •Entry guide: 05_Resources/network/index.md
  • •Requirements/architecture: 05_Resources/network/requirements-architecture.md
  • •Person card template: 05_Resources/network/people/_template.md
  • •CLI query tool (agent-friendly): 05_Resources/network/_cli/network.py
  • •CLI validator: 05_Resources/network/_cli/validate_people.py
  • •PRISM (portable ingest skill): 06_Infra/skills/portable/prism.md
  • •External source ingest notes (this skill): 06_Infra/skills/native/me-network-crm/references/external-source-ingest.md
  • •Fields/tags reference (this skill): 06_Infra/skills/native/me-network-crm/references/fields-and-tags.md

Workflow

A) Add / ingest a new contact (录入)

  1. •Choose a stable Handle/ID (ASCII, stable over time). Prefer primary-handle to make Ctrl+P/Quick Open easier.
  2. •Create 05_Resources/network/people/<handle>.md from 05_Resources/network/people/_template.md.
  3. •Fill the minimum fields for retrieval + action:
    • •Handle/ID, 关键词, Tags, 状态, 最后联系, 下次跟进 (or explicitly 暂无)
    • •Optional but useful for long-lived accuracy: 阶段锚点 (e.g. 2026-01-13（本科大三）)
    • •Optional for relative calculations: 出生日期 and/or 预计毕业
  4. •Convert raw context into searchable tokens:
    • •Put “domain + resource + scenario” into 关键词
    • •Add tags using the copy-paste list in 05_Resources/network/index.md and only add new tags when necessary (then update the list)
  5. •If the user provided a long summary, keep it in sections, but avoid sensitive PII and avoid storing raw passwords/keys.

B) Update an existing contact

  1. •Locate the card by tags/keywords/name:
    • •python3 05_Resources/network/_cli/network.py <keywords>
    • •python3 05_Resources/network/_cli/network.py --tag '#resource/intro' <keywords> --json
  2. •Append one line to 互动记录（流水） for the new event/outcome.
  3. •Update 最后联系 and set/adjust 下次跟进 date + action.
  4. •Adjust 状态 if needed (#status/active|warmup|dormant|archived).

C) Find people for a problem (检索匹配)

  1. •Translate the ask into tags + 2–5 keywords (domain + resource + scenario).
  2. •Query:
    • •python3 05_Resources/network/_cli/network.py --tag '#domain/ml' --tag '#resource/intro' <keywords> --json
    • •Time-aware stage inference (if cards have 阶段锚点): python3 05_Resources/network/_cli/network.py --as-of 2026-09-01 --json
    • •Relative to you (age / graduation deltas): python3 05_Resources/network/_cli/network.py --relative-to-me --json
      • •Uses 04_Assets/profile/Me.md by default (machine-readable fields); override via --me-path, --me-birth-date, --me-stage-anchor, --me-expected-graduation
      • •Needs contact 出生日期 and/or 预计毕业 (otherwise部分字段为空；预计毕业可由阶段锚点对本科/硕士做粗略推算)
  3. •Return top candidates with:
    • •Why they match (tags/keywords)
    • •Suggested outreach angle (1–2 sentences)
    • •Next action (update 下次跟进 in their cards if you’re taking action now)

Validation (optional but recommended)

Run:

  • •Preferred: python3 05_Resources/network/_cli/validate_people.py
  • •Legacy (this skill copy): python3 06_Infra/skills/native/me-network-crm/scripts/validate_people.py

Fix any missing required fields or malformed tags/dates it reports.

If python3 in your environment hangs (e.g. conda shim issues), retry with /usr/bin/python3.

D) Ingest external info sources (截图/聊天/简介页) with PRISM

Goal: turn “mixed raw context” into Facts / Inferences / Actions / Me-Info, then write it into 05_Resources/network/people/<handle>.md (append-only logs; update top fields as needed). If the user is sending a long timeline of chat screenshots, treat each screenshot as one ingest batch and keep the cumulative person card as the running log.

  1. •Identify the source type(s) and extraction constraints:
    • •Chat screenshot / chat text
    • •Profile page / bio / homepage
    • •Email / long-form note
    • •Mixed multi-source bundle (dedupe across sources)
    • •If content is missing or ambiguous, ask up to 3 minimal questions (name, relationship+strength, last contact date).
  2. •Extract and normalize raw signals into searchable tokens (don’t over-interpret):
    • •Pull explicit facts (names, school/company, stage, location, constraints, preferences)
    • •Keep uncertain items as “to confirm”
  3. •Run PRISM decomposition (follow the canonical prompt; keep Facts vs Inferences separated):
    • •Facts: only what is explicitly stated
    • •Inferences: cautious, with basis
    • •Actions: 1–3 dated follow-ups (date + purpose + tone)
    • •Me-Info: any “about me” signals in the content; prefer stable preference/boundary/strategy, but keep time-stamped notes even if they are small (the user wants complete records)
  4. •Store results:
    • •Update the card’s top bullets (关键词/Tags/最后联系/下次跟进/阶段锚点/预计毕业/出生日期 as applicable)
    • •Append one ## Prism（...） block (date-stamped) near the bottom; avoid sensitive secrets (passwords/keys)
    • •If you extract stable “about me” signals (preferences/boundaries/strategy), sync them into 04_Assets/profile/signals.md (create if missing; short, reusable bullets; time-stamped when possible)