Agent Ops Research Brief

FleetPack Skill · Pack #01 · v1.0.0
Install: drop this folder into .cursor/skills/ (or paste into a project rule / custom skill).
Brand: FleetPack — The Dude runs ops so you don't have to.

Purpose

Teach any agent how to run a weekly research → synthesis → one-page brief pipeline.
Output is a single, scannable ops brief a human can act on in under five minutes — not a wall of links.

Use this skill when the user asks for:

• A weekly / recurring research brief
• Competitive or landscape scan
• "What's new in X this week?"
• Source-backed synthesis with a clear recommendation

Non-goals

• Financial advice, investment tips, or crypto signals
• Scraping private / authenticated systems
• Dumping raw search results without synthesis
• Long reports (keep the deliverable to one page)

---

Pipeline (follow in order)

1. Scope lock (60 seconds)

Before searching, confirm or infer:

Field	Example
Topic	MCP marketplace landscape
Time window	Last 7 days (or stated range)
Audience	Founder / ops lead / eng lead
Decision	What should they do / watch / ignore?
Out of scope	Anything the brief must not cover

If the user gave a vague ask, state your assumed scope in one sentence and proceed. Do not stall.

2. Collect (web search + hygiene)

• Run 3–8 targeted searches (not one vague query). Mix: product names, "announcement", "pricing", "open source", year.
• Prefer primary sources: official blogs, docs, GitHub releases, reputable trade press.
• For each candidate source capture:
  • Title · URL · Publisher · Date · Claim in one line
• Source hygiene rules
  • Discard undated rumor threads unless corroborated by ≥2 independents
  • Flag paywalled or registration-gated items as [gated]
  • Never invent URLs, quotes, or dates
  • Prefer 2025–2026 material for "current landscape"; label older items as [background]

3. Scorecard (force ranking)

Score each finding 1–5 on three axes (write the numbers; don't just vibe):

Axis	Meaning
Signal	How much this changes the landscape
Confidence	Source quality + corroboration
Actionability	Can the reader do something this week?

Priority score = (Signal × 2) + Confidence + Actionability (max 15).
Keep the top 5–7 items for the brief; park the rest under "Parking lot" (one line each).

4. Synthesize

Write for a busy human:

• Lead with what changed and why it matters
• Group findings into 2–4 themes (not a chronological dump)
• Call out contradictions explicitly ("A says X; B says Y")
• Separate facts from FleetPack take (opinion)

5. Deliver (one-page brief)

Produce exactly the format in Deliverable format below.
Hard length target: ≤ 450 words for the main body (excluding source list).

---

Deliverable format

Copy this skeleton every time:

# Ops Brief: {Topic}
**Window:** {start} → {end} · **Prepared:** {date} · **Audience:** {role}

## TL;DR
{3 bullets max. Decision-oriented. No throat-clearing.}

## What moved
| # | Finding | Priority | So what? |
|---|---------|----------|----------|
| 1 | … | {score}/15 | … |
| 2 | … | … | … |
| 3 | … | … | … |

## Themes
### {Theme A}
{2–4 sentences + one concrete implication}

### {Theme B}
…

## Watch / Act / Ignore
| Bucket | Item | Why |
|--------|------|-----|
| **Act this week** | … | … |
| **Watch (30d)** | … | … |
| **Ignore for now** | … | … |

## FleetPack take
{4–6 sentences. Opinionated but honest. No hype.}

## Sources
1. [{Title}]({URL}) — {Publisher}, {Date} — {one-line claim}
2. …

## Parking lot
- {Low-priority note}

---

Quality bar (self-check before shipping)

• Scope stated; time window clear
• ≥3 independent sources cited with dates
• Scorecard numbers present (not just adjectives)
• Act / Watch / Ignore is non-empty and specific
• No secrets, no private URLs, no financial advice
• Main body fits one page / ≤450 words
• Tone: clear, dry-humor OK, never corporate sludge

---

Recurring / weekly mode

When asked to run this weekly:

1. Reuse last brief's Watch items as this week's seed queries
2. Diff: open with "Since last brief…" (2–3 bullets of deltas only)
3. Archive prior briefs as brief-YYYY-MM-DD.md if the user wants a trail

---

Voice (FleetPack / The Dude)

• Straight talk. Short sentences.
• "Here's what matters. Here's what doesn't."
• Light personality; zero condescension.
• Never name-drop unrelated products or client work.

---

Failure modes → fixes

Failure	Fix
Brief is a link dump	Cut to top 5 by priority; rewrite So-what column
Vague recommendations	Rewrite Act items as verbs + owners + timeframe
Stale sources	Re-search with current year; demote old to background
Scope creep	Move extras to Parking lot; protect the one-page limit

---

Install (Cursor)

1. Copy this entire SKILL.md into your project as
   .cursor/skills/agent-ops-research-brief/SKILL.md
   (or merge the pipeline into your team's custom instructions).
2. In chat: @agent-ops-research-brief (or invoke by name) and give a topic + window.
3. Optional: pin a recurring prompt —
   Run Agent Ops Research Brief for {topic}, last 7 days, audience = {role}.

That's the whole skill. Keep it tight. Ship the brief.