autopilot
The whole coding loop on one deliberate yes: map, spec, critic-hardened plan, GitHub issues, parallel worktree implementers, fresh-context review loops, one PR…
Morning work briefing — "what does my day look like, in one scan?" Gathers today's calendar and emails needing attention (via connected Google MCP tools when available), recent git activity and unfinished work across your local repos, your open tasks (TODO file / assigned GitHub
$ npx -y skills add duthaho/skillhub --skill daybrief --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/daybriefContext preview
The summary Claude sees to decide when to auto-load this skill.
Morning work briefing — "what does my day look like, in one scan?" Gathers today's calendar and emails needing attention (via connected Google MCP tools when available), recent git activity and unfinished work across your local repos, your open tasks (TODO file / assigned GitHub
name: daybrief description: >- Morning work briefing — "what does my day look like, in one scan?" Gathers today's calendar and emails needing attention (via connected Google MCP tools when available), recent git activity and unfinished work across your local repos, your open tasks (TODO file / assigned GitHub issues), and a quick radar check on topics you track — all in parallel — and renders one compact, prioritized brief in chat, carrying over unfinished focus items from yesterday's saved brief. Every source is optional and degrades gracefully; strictly read-only (never sends email, never touches the calendar, never pushes). Use at the start of a workday or when the user asks "what's on today", "morning brief", "daybrief", "catch me up", "what should I focus on today". For deep research on one topic use pulse instead.
`/daybrief [plain-language steering, e.g. "skip email", "just repos", "save it"]`
Answer one question in one scan: **what needs my attention today, in what order?** Pull every source in parallel, compress hard, and lead with the priorities — not the raw data. Target: the user reads it in under two minutes.
Look for `daybrief-config.md` in the current working directory. It defines:
# daybrief config ## Repos # absolute paths, or workspace roots to scan one level deep - /home/hop/code ## Tasks # optional: where your todos live (any/all of these) - file: TODO.md # a local markdown todo file (path relative to a repo or absolute) - github: assigned # issues assigned to you, via `gh` if authenticated ## Radar topics # 2–4 topics to check for movement - claude code - postgres ## Preferences - timezone: Asia/Ho_Chi_Minh - email: on # on / off - calendar: on # on / off - save: off # on = also write brief to out/daybrief/ (enables carry-over)
If missing, **don't block**: run with what's inferable (current repo, no radar topics). Then, **after** delivering the brief, offer to create the config via a short interview — ask (in one compact question set, prefilled with what you learned this run): which repos/workspace roots to watch, 2–4 radar topics, where todos live (file path / GitHub / none), email & calendar on/off, and whether to save briefs (`save: on` enables carry-over). Confirm the timezone from the system rather than asking. Write `daybrief-config.md` from the answers and remind that it's gitignored. If the user declines, don't ask again on later runs — just note "no config" in the footer. Honor one-off steering from the arguments over the config.
Launch everything concurrently. Each source is independent and optional — if its tools aren't connected or it errors, mark it "unavailable" in the brief footer and move on. Never let one dead source stall the brief.
With five optional sources it's easy to silently drop one. Track this checklist through the run — the brief isn't done until every box is ✓ or a conscious ✗ with the reason shown in the footer (a skip is a decision, not an oversight):
- [ ] calendar - [ ] email - [ ] repos - [ ] tasks - [ ] radar - [ ] carry-over judged (if save: on) - [ ] Focus = top 3, each with a "when"
available; otherwise skip.
link. Flag: the next event and minutes until it, overlaps, meetings with no agenda/description (prep risk), and the largest free block (deep-work slot).
(direct questions/requests to the user), **FYI worth knowing**, and count the rest as noise (report the number only). Quote nothing sensitive at length — one line per thread, sender + ask.
commits since yesterday morning (`git log --since`), dirty working trees (`git status --short`), unpushed branches (`git log @{u}..` where an upstream exists), and — if `gh` is authenticated — open PRs awaiting the user's review or with fresh comments.
today's loose ends.
boxes / non-struck lines), top ~5 by position. Read-only; never reorder or check anything off.
--state open` across the configured repos (plus `gh search issues` for the rest), top ~5 by recency. Skip silently if `gh` is missing/unauthenticated.
from git activity.
topic + one web search pass. Return only items with real traction (points/ comments) — **max 2–3 items total across all topics**, each one line + link. This is a glance, not a pulse; suggest `/pulse <topic>` if something is hot.
**Carry-over (if `save: on`):** before ranking, read the most recent brief in `out/daybrief/`. For each of its Focus items, judge from today's evidence whether it happened — commits/pushes touching it, the PR merged, the meeting past — and sort into **done** vs **carried over**. Carried-over items start near the top of today's ranking (unfinished yesterday-priorities outrank new shiny things), and the brief reports the follow-through score ("yesterday: 2/3 done"
Give Claude Code a memory and make it cite its sources — skills for research, daily work, and shipping code. They started as prompts I kept retyping, so I wrote each one down once. Keyless: no API keys, no signups.
The whole coding loop on one deliberate yes: map, spec, critic-hardened plan, GitHub issues, parallel worktree implementers, fresh-context review loops, one PR…
Human-facing documentation, architecture docs, and mermaid diagrams for a codebase, a module, or a feature — "draw me the architecture, with receipts." Fans…
Lightweight bug-fixing loop — reproduce → root-cause → fix test-first → verify — deliberately separate from the heavyweight feature workflow so small fixes…
Prompt-cache economics forensics — why your Claude Code tokens burn so fast, with the numbers to prove it. Parses local ~/.claude transcripts (0 tokens, a…
End-of-session shipping gate — "prove it works, then ship it." Runs an evidence checklist (full test suite, lint/typecheck, build, and actually running the…