# Gh Weld Activity

> GitHub activity digest — lists issues and PRs grouped by time window (today, last 3 days, last 7 days, last 30 days). Supports repo-scoped or GitHub-wide views, selected by single keypress. Use when: the user wants a repo overview, activity summary, recent activity digest, or to see what's changed. Trigger phrases: "what's been happening", "activity digest", "repo overview", "recent activity", "what's new".

- Skill: `wrathza/gh-weld-activity` (Agent Skill)
- Install (CLI): `npx skillmds@latest add wrathza/gh-weld-activity`
- Raw SKILL.md: https://api.skillmd.com/api/skills/wrathza/gh-weld-activity/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: WrathZA (https://skillmd.com/u/wrathza)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/wrathza/gh-weld-activity

---


# gh-weld-activity

Activity digest grouped by recency. Fetch, bucket, render.

## NEVER

- **NEVER omit a time window because it has no activity**
  **Instead:** Show a `None` placeholder under Issues and PRs.
  **Why:** Omitting windows makes the digest look misleading — the user can't tell if nothing happened or the window was skipped.

- **NEVER use `|` (pipe) in Bash calls**
  **Instead:** Read JSON output directly from the tool result — no piping needed in this skill.
  **Why:** Claude Code stops execution when it encounters a pipe, interrupting the session without warning.
  Note: `|` in markdown table row syntax is unaffected.

- **NEVER chain Bash calls with `&&` or `;`**
  **Instead:** Run each command as a separate Bash tool call.
  **Why:** Claude Code's safety check fires on ambiguous multi-command calls and interrupts mid-flow.

- **NEVER use `$()` command substitution**
  **Instead:** Read command output directly from the Bash tool result.
  **Why:** Claude Code's permission system prompts on `$()` during execution, interrupting unnecessarily.

- **NEVER mix items across windows**
  **Instead:** Each item appears in exactly one window — the most recent it qualifies for.
  **Why:** Overlapping windows inflate counts and confuse the user about what's actually new.

## Workflow

### 1 — Scope

Ask: `Scope: (r)epo or (g)ithub-wide? [r]`

- `r` or Enter → current repo only
- `g` → all repos visible to the authenticated user

If the response is neither `r`, `g`, nor Enter: re-prompt once, then default to repo scope.

### 2 — Pre-flight

```bash
gh auth status
```

If this fails: output "gh is not authenticated — run `gh auth login` first." and stop.

Check gh version:

```bash
gh --version
```

`gh search` requires gh v2.29+. If the version is older and the user chose GitHub-wide scope, warn: "gh search requires v2.29 or later — falling back to repo scope." and proceed as repo scope.

### 3 — Fetch

**Repo scope:**

```bash
gh issue list --state all --limit 200 --json number,title,state,author,url,createdAt
```

```bash
gh pr list --state all --limit 200 --json number,title,state,author,url,createdAt
```

**GitHub-wide scope:**

```bash
gh search issues --limit 200 --sort created --json number,title,state,author,url,createdAt,repository
```

```bash
gh search prs --limit 200 --sort created --json number,title,state,author,url,createdAt,repository
```

Read the JSON from each tool result directly — no temp files needed. The `author` field is an object: use `author.login` for the display name.

If any fetch returns exactly 200 items, it likely hit the `--limit` cap: note "⚠ Results capped at 200 — older items beyond the cap are omitted." in the digest header so the view doesn't read as complete.

If a fetch fails (rate limit, network error — most likely on GitHub-wide `gh search`): output "Fetch failed: `<gh error>` — try again shortly, or narrow to repo scope." and stop. If a fetch succeeds but returns an empty list, that is valid — proceed and render `None` placeholders rather than treating it as an error.

> Discussions are not supported by the `gh` CLI JSON output; omit silently.

### 4 — Bucket

Use today's date from the session context (the current date is provided to you — do not shell out to `date`, which would need `$()` to capture). Treat it as UTC. Classify each item into exactly one window:

| Window | Item's `createdAt` is on or after… |
|--------|-------------------------------------|
| Today | today 00:00 UTC |
| Last 3 days | today − 2 days 00:00 UTC |
| Last 7 days | today − 6 days 00:00 UTC |
| Last 30 days | today − 29 days 00:00 UTC |

Assign to the most recent qualifying window. Items older than 30 days are omitted.

Windows are computed in UTC, so "Today" means the current UTC day — an item created late in the user's local evening may land in "Today" or the previous window depending on the UTC offset. State "(windows are UTC)" once in the digest header so counts near midnight aren't misread.

### 5 — Render

```
## Today
### Issues
- #42 Fix login crash [open] — @alice — https://github.com/...
### Pull Requests
- #43 Add dark mode [merged] — @bob — https://github.com/...

## Last 3 days
### Issues
None
### Pull Requests
- #41 Bump deps [closed] — @carol — https://github.com/...

## Last 7 days
...

## Last 30 days
...
```

For **GitHub-wide scope**, prefix each item with the repo: `owner/repo#42 Title [state] — @author — url`.

Show item counts per window as a header suffix: `## Today (3 issues, 1 PR)`.