thread-tracker

Thread Tracker Skill

Track and summarise Slack threads locally inside an agent's assigned working folder.

---

Storage Layout

All files live relative to the agent's assigned folder (the folder the cowork agent has been given as its workspace).

<assigned-folder>/
├── Threads.txt          # Human-readable index of all tracked threads
└── threads/             # One .md file per thread
    └── <filename>.md

Threads.txt format

Each entry is a YAML-like block separated by blank lines:

- url: <slack_thread_url>
  file: <filename>
  channel: <channel_name>
  thread_id: <thread_ts>
  first_message_summary: <one-sentence summary of the opening message>
  last_read_time: <ISO 8601>

Thread filename format

<channel>__<3-word-context>__<@person1-@person2>__<YYYY-MM-DD>.md

• channel — extracted from the Slack URL (e.g. C01AB2CD3EF → use the human-readable channel name if discoverable, otherwise the raw ID)
• 3-word-context — three hyphen-joined lowercase words summarising the thread topic (generated after reading the first few messages)
• @person1-@person2 — @handle of the most active/tagged people in the first few messages, up to 3 handles joined with -; omit the @ in the filename itself (e.g. alice-bob-carol)
• YYYY-MM-DD — date of the first message in the thread

Example: general__deploy-pipeline-broken__alice-bob__2024-03-15.md

---

Thread File Template

## Info
- Thread link: <url>
- Channel: <channel name>
- Start time: <ISO 8601 of first message>
- Last-read-time: <ISO 8601>
- Context: <one-sentence description of what the thread is about>
- People involved: <@handle1>, <@handle2>, ...
- Project details: <project/team name if discernible, else "N/A">

## Content
<!-- Bullet-point summary of the discussion. Attribution format: **@handle** — point made. -->

## References
| Link | Shared by | Context |
|------|-----------|---------|

## Impactful
<!-- Who drove the conversation most — ranked list with brief reasoning. -->

## Conclusion
<!-- Only written if the last reply is > 1 month ago. Otherwise this entire section is OMITTED. -->
<!-- Summary of outcome/resolution. -->

---

Implementation Notes

• Always use the Slack MCP tool (e.g. mcp__claude_ai_Slack__slack_read_thread) to fetch thread content. Never ask the user to paste thread content manually.
• When spawning Explore agents in parallel, pass each agent the full thread URL, the assigned folder path, and the existing Threads.txt content so it has all necessary context.
• ISO 8601 timestamps should include timezone offset, e.g. 2024-03-15T14:32:00+05:30.
• Handle Slack rate-limit errors gracefully: if an agent fails to fetch a thread, log the error in the thread file under a ## Errors section and continue processing other threads.
• Never overwrite a thread file from scratch on update — always merge to preserve historical Content bullets and References rows.