Slack Kb Channel

---
name: slack-kb-channel
description: >-
  Build or update a Slack channel knowledge base for a given month. Works for
  any accessible channel given its Slack channel id. Ingests threads via the
  Slack API and produces temporal monthly memory files.
---

# Slack channel knowledge base

Use when building or updating a **Slack channel** knowledge base for a given month. Works for any channel you have access to; you need its **Slack channel id** (starts with `C` for public channels).

## Memory (canonical — ship this)

**Temporal memory** (time-sliced, one file per calendar month) lives under **`memory/temporal/`**. That directory name signals content keyed by **when** it was observed, not by a long-lived theme.

| Path | Purpose |
|------|---------|
| `memory/temporal/<channel-slug>-YYYY-MM.md` | Channel themes, incidents, pointers, and **thread-derived findings** (thematic digest keyed by `thread_ts`; no exhaustive thread list in the file) |

Use a **stable short slug** for the channel in the filename (e.g. `help-ml-infrastructure`, `team-foo-oncall`), so the same channel always maps to the same prefix. You may include the channel id in prose or front matter if that helps disambiguation.

Do **not** add a separate `-threads.md` or other companion knowledge base files for the same month.

**Rules for memory**

- **Dates in prose:** Prefer **`YYYY/MM/DD`** for a specific day and **`YYYY/MM`** for a month (e.g. incident windows, “seen in” notes). Avoid English month names or abbreviations (`Feb`, `March 2026`) in running text unless quoting Slack verbatim.
- Do **not** include a **thread index** (full table of every `thread_ts` + topic) in the monthly markdown. Use **thread-derived findings** only, with `thread_ts` inline where it helps.
- Do **not** link to, name, or point readers at **raw thread capture files** (e.g. per-thread markdown exports, “dump” directories, or filenames derived from timestamps). Those are ingestion artifacts, not the knowledge base.
- Thread identity in memory is **`thread_ts`** (and **channel id** when the slug alone is ambiguous). People re-open discussions in Slack using those ids.
- When merging new facts from Slack, update the single monthly file under `memory/temporal/` only; do not add “see file X in folder Y” for raw captures.

**Theme-organized views** (merged across months, grouped by topic) are maintained separately — see the [`kb-organized-memory`](../kb-organized-memory/SKILL.md) skill.

---

## Ingestion (maintainers only — not part of memory)

[`build_thread_kb.py`](build_thread_kb.py) pulls history and threads directly via the Slack API. Credentials (`SLACK_TOKEN`, `SLACK_COOKIE`) are loaded from `.env`. Raw outputs may be stored outside `memory/temporal/` or in tooling-only paths; **keep them out of the canonical temporal memory docs.**

### Batch thread metadata (recommended)

[`build_thread_kb.py`](build_thread_kb.py) walks **any channel** (by Slack channel id) using `conversations.history` + `conversations.replies` directly. It supports a full **calendar month** (`--month`) or an arbitrary **date range** (`--after`/`--before`). It writes **JSONL** (one object per thread: `channel`, `ts`, `replies`, `preview`) — a **maintainer artifact** for ingestion and validation; it is not the published knowledge base and is not pasted into memory as a thread table.

Pass **`--channel <CHANNEL_ID>`** or set **`SLACK_CHANNEL_ID`**. Default output files include the channel id so different channels for the same period do not overwrite each other.

```bash
# Full calendar month
uv run skills/slack-kb-channel/build_thread_kb.py --channel <CHANNEL_ID> --month 2026-03 \
  --output memory/temporal/thread-fetch-2026-03-<CHANNEL_ID>.jsonl \
  --write-valid-list

# Arbitrary date range (e.g. a week)
uv run skills/slack-kb-channel/build_thread_kb.py --channel <CHANNEL_ID> \
  --after 2026-04-07 --before 2026-04-14 \
  --output /tmp/channel-week.jsonl
```

- **`--skip-threads`**: only print / list root `thread_ts` values in the range (no per-thread API calls).
- **`--write-valid-list`**: also writes `memory/temporal/extracted-thread-ts-<date>-<CHANNEL_ID>-valid.txt` (excludes rows where the thread fetch failed).

**Workflow**

1. Note the channel id and a **stable `<channel-slug>`** for filenames. Run **`build_thread_kb.py`** for `YYYY-MM` (or the relevant date range).
2. Treat reply permalinks as **non-parent** timestamps when scraping; the script uses channel roots only.
3. Optional: commit or ignore `memory/temporal/thread-fetch-*.jsonl` and `memory/temporal/extracted-thread-ts-*-valid.txt` per your hygiene.
4. Refresh **`memory/temporal/<channel-slug>-YYYY-MM.md`**: themes and **Thread-derived findings** (with `thread_ts` where useful). Do **not** add a full thread index table to the memory file.
5. When themes should span months, refresh the organized output per [`kb-organized-memory`](../kb-organized-memory/SKILL.md).