Office Collaboration Radar — English

Turn scattered collaboration records into a stable, auditable project status card. Extract only facts supported by the supplied material. Do not produce a full weekly report, judge individual performance, send reminders, or write to external systems.

Language choice: use this package only after the user requests English analysis or output. For Chinese material, use Office Collaboration Radar.

Capabilities

• Produce seven fixed modules: overview, progress, decisions, actions, risks and dependencies, cross-functional relationships, and human-review items.
• Attach evidence snippets of at most 80 characters; use Not provided when no reliable evidence exists.
• Redact phone numbers, email addresses, government IDs, bank-card numbers, common API-key and token forms, authorization headers, passwords, private-key blocks, and user-supplied entity names.
• Flag conflicting owners, deadlines, or statuses for human review.
• Rank actions as P0/P1/P2, synthesize an executive summary, and generate an SVG collaboration-health radar.
• Export the structured card to CSV, Feishu/Bitable mapping JSON, or Notion mapping JSON.

Input

Accept Markdown, plain text, exported chats, meeting notes, or project logs. Label the source of each segment when combining materials. Optional metadata includes project name, time range, departments, and a list of known entities to redact.

If the material does not state a project name, time range, owner, deadline, or department, use Not provided. Do not infer or complete it.

Workflow

1. Separate facts from opinions, questions, tentative statements, small talk, and duplicates.

2. Draft the seven-module card using the key order in templates/json-output-schema.md.

3. Keep owner and deadline values verbatim. If an exact value cannot be found in the supplied source, replace it with Not provided.

4. Enforce evidence, redaction, Markdown-cell sanitization, conflict checks, and priority ordering:

   python3 scripts/process.py enforce --card draft-card.json --source source.txt --out enforced-card.json
   

   Aggregate multiple cards first when needed:

   python3 scripts/process.py aggregate --cards card-1.json card-2.json --out aggregated.json
   

5. Optionally generate a radar chart:

   python3 scripts/radar_chart.py generate --card enforced-card.json --out collaboration-radar.svg
   

6. Optionally export a downstream mapping:

   python3 scripts/export_card.py export-card --card enforced-card.json --format csv --out actions.csv
   

7. Return the Markdown card, JSON result, and executive summary. Before publishing or changing the skill, run:

   python3 scripts/process.py selftest
   

All file-producing commands require an explicit --out path and refuse to replace an existing file unless the user deliberately adds --force. Never reuse an input path as an output path.

Local access boundary

• Read only the exact --card, --source, --entities, or --cards paths supplied by the user. Do not enumerate directories or search for other files.
• Do not read environment variables, credential stores, browser data, shell history, or unrelated workspace content.
• Do not use the network or transmit source material. This skill has no MCP or external-service dependency.
• Write only to an explicit --out path. The default export behavior is stdout; every other file-producing command requires --out.
• Treat --force as explicit permission to replace only the named output file. Never infer overwrite permission.

Output contract

Keep every heading in this order, even when a section contains only Not provided:

# Collaboration Status Card

## Project Overview
## Progress
## Confirmed Decisions
## Owner × Deadline Actions
## Risks / Blockers / Dependencies
## Cross-functional Relationships
## Human Review Required
## JSON Output

Use templates/collaboration-status-card.md for the Markdown layout and templates/json-output-schema.md for the machine-readable contract.

Evidence and safety boundaries

• Do not turn tentative intent into a commitment or invent owners, deadlines, decisions, departments, or launch dates.
• Use Conflict detected; human review required consistently and add a matching human-review item.
• Sanitize all user-derived Markdown table values and escape CSV formula prefixes.
• Escape SVG titles before rendering them.
• Recompute every evidence snippet from the supplied source; never trust a prefilled evidence field.
• Redact common credential patterns, but do not treat the deterministic redactor as a complete secret scanner. Ask the user to pre-redact live credentials, customer data, and sensitive HR material.
• Do not send messages, email, or calendar invitations, and do not modify Jira, Feishu, Notion, CRM, calendars, or other systems.
• Do not convert interpersonal friction or emotional language into performance judgments or blame.

The example under examples/enterprise-ai-pilot/ is synthetic and redacted. It demonstrates extraction behavior only; it is not evidence of a production deployment or business outcome.