Me Network CRM

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
• Skill-local copy: python3 .agents/skills/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)