automation-scout

name

automation-scout

description

Use this skill when a user asks what parts of their workflow could be automated, what repetitive work they are doing, what agents/bots/skills/workflows/AI solutions would save them time. It scans recent workplace signals across configured sources — existing cards/briefs, Slack (including team channels and DMs), WorkIQ/Microsoft 365, emails, calendar events, and GitHub activity (including owned repos) — to produce concrete automation and AI opportunities for personal improvement. Reads full message bodies, emails, and documents to find the best opportunities; never writes to Slack/email/GitHub. Output is a ranked scout brief with evidence, time-saved estimate, and next action.

automation-scout

Find repetitive work the user is already doing and turn it into concrete automation candidates.

The skill answers one question:

What work should I stop doing manually?

It is a discovery skill, not an execution skill. It recommends automations; it does not build, schedule, send, label, comment, or mutate anything.

Positioning

This is a deep-reading automation recommendation skill tailored for personal improvement and individual knowledge workers.

It differs from workflow builders because it starts with workflow discovery, not "build me a Zap." It looks for repeated patterns across collaboration and development surfaces (reading full context, messages, and emails), then suggests small automations the user could delegate to Copilot, a bot, a scheduled workflow, an AI solution, or a reusable skill.

Prefer non-SWE fixes first. The default recommendation ladder is:

1. Stop doing the work.
2. Batch or route it with existing product settings.
3. Use no-code or low-code automation in existing tools.
4. Use an AI solution (like Copilot skills, Copilot Workspace agents, or reusable prompts) with human review.
5. Build custom software only when the first four options cannot handle the toil.

Success criteria

Produce 3-7 useful automation candidates where each candidate has:

1. A clear repetitive-work pattern.
2. Evidence from at least one source.
3. A concrete automation shape.
4. A privacy level and data needed.
5. A realistic next action.

If data is thin, produce a sparse scout brief. Do not pad with guesses.

Context Gathering

Use full context and content to find the best automation candidates:

• Slack: team channels, DMs, thread/reaction counts, file/link presence, and full message bodies.
• Email: sender, timestamp, subject lines, and full body text.
• Calendar/WorkIQ: event title, time, attendees, meeting bodies, and linked-doc content.
• GitHub: owned repos, watched repos, issue/PR number, title, labels, assignees, and full issue/PR bodies.
• Existing cards/briefs: signal reason, source refs, timestamps, and full source bodies.

Source order

Use the cheapest available source first. Stop when there is enough signal for a useful brief.

1. Existing cards, briefs, indexes, or prior scout outputs in the configured workspace.
2. Existing automation inventory:
   • configured local skills
   • configured workflow prompts
   • configured scheduled automations Mark already-covered patterns as "existing automation to tune," not "new automation to build."
3. GitHub data (including bodies) for owned repos, watched repos, recent authored/assigned/review-requested items.
4. Calendar/email full context through WorkIQ.
5. Slack allowlisted-channel and DM full message content.

If a source is unavailable, note it under Sparse areas and continue.

Config and gates

Prefer configured workspace settings:

• source allowlists
• retention rules
• automation inventory

When querying WorkIQ for email metadata, you can include full email subject lines, body text, and attachments to get the best automation signals.

Lookback

Default lookback is 30 days. Use 14 days for a quick scout and 90 days only when the user asks for deeper trend detection.

Detection patterns

Score each pattern by frequency, friction, automatable shape, privacy risk, and user leverage.

1. Repeated status synthesis

Signals:

• Recurring status/update meetings.
• Repeated Slack or GitHub threads about the same project.
• Repeated asks like "what's the status," "weekly update," "where are we on."
• Manual roundups across repos, Slack channels, docs, or email.

Automation shapes:

• Scheduled weekly brief.
• Project pulse digest.
• GitHub issue/PR + Slack signal rollup.
• Exec-ready summary draft.

2. Repeated meeting prep

Signals:

• Recurring 1:1s or project syncs.
• Same attendee set.
• Recent activity clustered around the meeting topic.
• Prior open loops that reappear.

Automation shapes:

• 1:1 prep packet.
• Meeting brief with open loops and suggested questions.
• Pre-read finder.

3. Follow-up debt

Signals:

• "I'll follow up," "circling back," "can you send," "reminder," "nudge."
• Threads where the user is asked a question and no reply/reaction is visible.
• GitHub review requests older than 24-72h.
• Email threads with follow-ups and no user reply.

Automation shapes:

• Follow-up tracker.
• Daily owed-reply queue.
• Review-request reminder.
• Draft-only nudge generator.

4. Repeated GitHub triage

Signals:

• Similar labels, titles, assignees, or comments across issues/PRs.
• Repeated issue routing or review assignment.
• Repeated PR review checklist.
• Same repo questions coming up across channels.

Automation shapes:

• Triage helper skill.
• Label suggestion draft.
• Review checklist generator.
• Issue duplicate finder.

5. Approval or review chasing

Signals:

• Recurring approval language.
• Long-lived review requests.
• Same approver set.
• Calendar/email reminders for approvals.

Automation shapes:

• Approval tracker.
• Review SLA dashboard.
• Draft-only reminder.

6. Repeated copy/paste reporting

Signals:

• Same sources repeatedly summarized into docs, emails, GitHub issues, or Slack posts.
• Calendar blocks named "write update," "weekly report," "digest," or "status."
• GitHub activity followed by Slack/email reporting.

Automation shapes:

• Report generator.
• Weekly digest.
• Draft status update.

7. Search/research loops

Signals:

• Repeated searches for the same repos, docs, channels, people, or decisions.
• Questions that ask "where is," "who owns," "what did we decide," "find the latest."

Automation shapes:

• Search macro.
• Decision index.
• Source-specific lookup skill.

8. Notification compression

Signals:

• High-volume system notifications from the same service.
• Repeated GitHub Actions, Dependabot, bot, workflow, alert, or notification metadata.
• Time-of-day notification bursts.
• Similar labels or titles indicating failures, partial runs, stale jobs, or missing data.

Automation shapes:

• Morning digest.
• Repo/service grouped notification rollup.
• State-change-only alerts.
• Failure-mode summary.

9. Recurring data/report assembly

Signals:

• Repeated Kusto, Trino, Unwrap, Power BI, or GitHub data questions.
• Calendar blocks or GitHub issues named weekly, monthly, baseline, report, digest, metrics, or trend.
• Same query-backed numbers repeatedly copied into briefs, docs, issues, or decks.

Automation shapes:

• Metric card generator.
• Weekly data brief.
• Saved-query runner with source refs and caveats.
• Draft report from metric cards.

10. Scheduling thrash

Signals:

• Repeated reschedule, find-a-time, hold, conflict, tentative, or "ask before booking" calendar metadata.
• High attendee-count meetings with low participation need.
• Self-blocks repeatedly moved or overridden.

Automation shapes:

• Focus-time guardrail.
• Optional-meeting triage.
• Reschedule assistant draft.

11. Repeated-answer FAQ

Signals:

• Repeated questions about the same policy, process, ownership, repo, channel, or decision.
• Same docs or issues repeatedly cited in Slack, email, or GitHub.
• Search/research loops that end at the same source.

Automation shapes:

• FAQ bot answer card.
• Team bot knowledge entry.
• Search macro with citations.

Ranking rubric

Score each candidate 1-5 in these dimensions:

• Frequency: How often the pattern appears.
• Friction: How much manual work/context switching it creates.
• Automatability: Whether inputs, outputs, and cadence are clear.
• Leverage: Whether it helps the user or a larger group.

Rank by:

frequency + friction + automatability + leverage

Break ties by choosing the smaller, earlier-win automation.

Scout brief shape

Return this exact shape:

# Automation Scout Brief: <lookback>

## Top opportunities

### 1. <short name>

**Pattern:** <what appears repetitive>
**Evidence:** <2-4 cited bullets; full content citations are encouraged>
**Suggested automation:** <specific agent/bot/workflow/skill/AI solution>
**Non-SWE path:** <settings/process/no-code/AI solution/prompt-first option>
**Data needed:** <sources used>
**Estimated time saved:** <rough range; mark as estimate>
**Why this first:** <one sentence>
**Next action:** <smallest useful step>

## Lower-confidence candidates

- <candidate> — why lower confidence / what would confirm it

## Sparse areas

- <source unavailable or intentionally skipped>

## Not recommended

- <pattern intentionally not recommended, especially because of privacy/sensitivity>

Citation rules

• GitHub evidence must cite an issue/PR/repo URL.
• Slack evidence must cite a Slack permalink or slack:// URL. Include quoted message bodies to provide context.
• WorkIQ evidence must cite the conversation id, meeting title/date, or file URL. Include private body text, email subject lines, or inferred email subject patterns to support the automation candidate.
• Existing cards or briefs cite their file path and source refs.
• Aggregate evidence may cite the source query/run plus aggregate/derived when no single item citation exists. Do not invent item-level citations for aggregates.
• Estimates must be labeled as estimates.

Recommendation ladder

Every top opportunity should include the lightest viable path:

• Stop: unsubscribe, close stale item, remove notification, cancel meeting, delete duplicate, mark no-action.
• Route: email rule, GitHub saved search, notification filter, project view, label view, calendar category.
• Batch: digest, weekly queue, office-hours review, triage block, recurring checklist.
• No-code/low-code: Power Automate, Slack Workflow Builder, GitHub Projects views, Outlook rules, calendar rules.
• AI Solution/Skill: reusable Copilot prompt, Workspace agent, or CLI skill that drafts/recommends but does not write externally.
• Custom SWE: last resort only.

Time-saved estimate

Use a coarse estimate unless actual durations are available:

• Small: under 15 minutes/week
• Medium: 15-60 minutes/week
• Large: over 60 minutes/week

If you give a numeric range, state the method: frequency x rough_minutes_per_instance, labeled as an estimate.

Output tone

Be direct and practical. Avoid pitch language. Do not say "transform," "unlock," "revolutionize," or "game-changing." The user needs a useful list, not a product page.

What this skill does NOT do

• Send Slack messages, emails, or meeting invites (it is read-only).
• Comment on, label, assign, close, or merge GitHub issues/PRs.