Appendix F: The Skill Library

Appendix F · Appendices

These are written to be copied and run, not admired. Each one is a skill: a document a model follows. Paste one into whatever AI tool your team uses, replace the bracketed parts with your own sources and rules, and run it. Then rewrite it every time it gets something wrong, because the first version will.

Every skill here ends with a human decision. Keep it that way.


F1. The Monday Review

NAME: monday-review
WHEN: Every Monday before 8:00 AM.

GOAL
Give me a proactive read on where we are against our goals, how spend is
pacing, and what we should be doing that nobody has asked for yet.

SOURCES (read all of them; say which ones you could not reach)
1. [CRM]: pipeline by stage, deals created and closed in the last 7 days,
   deals with no activity in 21 days.
2. [Internal search / wiki]: anything published or changed in the last 7 days
   about customers, competitors, launches or plans.
3. [Docs folder]: this quarter's goals and budget.
4. [Last week's review], so you can say what changed.

STEPS
1. State this quarter's goals in one line each, from the goals doc.
2. For each goal: on pace, at risk, or off pace, with the number and its source.
3. Spend: planned vs actual to date, by program. Flag anything more than
   15% off plan in either direction.
4. What changed since last Monday that matters. Maximum five items.
5. Opportunities nobody asked for: up to three things the data suggests we
   should build, stop or change. For each, the evidence and the smallest test.
6. Questions I should ask this week, and who to ask.

RULES
- Every number needs a source. If you can't source it, say so; don't estimate.
- Separate what the data says from what you infer. Label inferences.
- Assume the CRM has stale stages and optimistic deals. Note anything that
  looks like it.
- No more than one page.

OUTPUT
One page: Goals / Spend / What changed / Opportunities / Questions.

HUMAN DECISION
I decide which opportunities to pursue. You propose; you don't assign work.

F2. The Voice Profile Builder

NAME: voice-profile-builder
WHEN: Once per person, then updated every time they correct a draft.

GOAL
Build a profile that lets a model draft in [PERSON]'s voice well enough that
they edit rather than rewrite.

SOURCES (in order of value)
1. Their corrections to drafts: the before and after of every sentence they
   rewrote. This is the most important input.
2. Unpolished writing: chat messages, quick emails, comments.
3. Transcripts of them talking: podcasts, talks, recorded calls.
4. Published writing LAST, because someone else probably edited it.

PROFILE SECTIONS
1. Register: how formal, how long the sentences run, how they open and close.
2. Always: the moves they make constantly.
3. Never: words, phrases and structures they never use or always remove.
4. Punctuation: how they actually punctuate, including what they avoid.
5. Point of view: the positions they hold and repeat.
6. Imitation tells: the things a copycat overdoes. Signature phrases, used
   sparingly. Note how often each one really appears.
7. Ten example sentences, verbatim, that sound most like them.

RULES
- Most of a real voice is ordinary sentences. Say so in the profile.
- Every rule needs an example from the sources.

HUMAN DECISION
The person reads the profile and corrects it. Their corrections go in.

F3. The Weekly Content Scout

NAME: weekly-content-scout
WHEN: Every Monday.

GOAL
Find the stories this week that [PERSON] has a real point of view on, and
propose a short plan. Do not summarize the news.

SOURCES
1. [List of publications and feeds]. Include at least three that competitors
   probably don't read.
2. [PERSON]'s voice profile and their last ten posts.

STEPS
1. Collect the 20 most relevant stories of the last 7 days.
2. Apply the protagonist filter to each: can [PERSON] speak to this from
   something they've actually done or seen? If they'd only be repeating the
   story, drop it.
3. Rank what survives. Rank the strange story above the big one if both pass.
4. Pick two or three. For each: the story, the angle in one sentence, why it's
   theirs to say, and the obvious take everyone else will post (so we avoid it).
5. Draft each post against the voice profile.

RULES
- Use the forbidden-words list: [LIST].
- No post may be a summary of the article.

HUMAN DECISION
[PERSON] picks, edits and posts. Nothing goes out automatically.

F4. The First-Meeting Brief

NAME: first-meeting-brief
WHEN: The evening before every first meeting.

GOAL
The rep walks in already knowing who's in the room and why they might care.

SOURCES
1. [CRM]: history with this account and these people.
2. Attendees' public professional profiles: role, tenure, previous companies.
3. What attendees have posted or said publicly in the last 90 days.
4. Company news from the last quarter.
5. Visible changes to their technology stack.
6. Any public complaints about their current vendor.
7. [Competitive library] entry for their current vendor.

OUTPUT (one page, five sections)
1. Who's in the room, one line each.
2. What they probably care about this quarter, with evidence.
3. The one thing they're likely to push back on.
4. The question to ask in the first five minutes.
5. What not to say.

RULES
- Every claim has a source link. Mark anything you're unsure of.
- Check that names, titles and employers match across sources. If they
  conflict, say so.
- This is for the rep's understanding, not for reciting to the buyer.
- If this brief would read the same for a different company, it's too generic.
  Rewrite it.

HUMAN DECISION
The rep decides what to use and verifies anything marked unsure.

F5. The Anti-Fluency Pass

NAME: anti-fluency-pass
WHEN: On any draft before it goes to review.

GOAL
Find where the draft sounds like the rest of the category, and fix it.

STEPS
1. Write the most typical version of this piece for a company in [CATEGORY].
   Label it THE MEAN.
2. Compare the draft to THE MEAN. List every sentence that a competitor could
   publish unchanged.
3. Flag every word from the forbidden list: [LIST].
4. Flag every comparative: faster, smarter, better, more, easier.
5. As a skeptical buyer who has read eleven competitor pages this morning,
   name the three sentences you'd skip and the one you'd remember.
6. For each flagged sentence, propose a rewrite that uses something specific:
   a number we can source, a customer's words, a concrete image, or a claim
   someone could disagree with.

RULES
- Don't make it smoother. Smoother is the problem.
- Don't invent facts to make a sentence specific. Ask for them instead.

HUMAN DECISION
The author decides which rewrites to take.