Agent skills for discovery notes and statements of work
A statement of work skill takes discovery notes and produces the document that decides whether the engagement goes well: deliverables, assumptions, exclusions, acceptance criteria and a timeline. The hard part of a SOW is not writing it, it is noticing what the notes did not say. A good skill is built to surface those gaps as questions rather than fill them in.
The gap list is the point
Every discovery call leaves things unsaid, and the unsaid things are what get argued about later. A skill that quietly invents an assumption is worse than no skill, because the invention arrives in confident, formatted prose. Build the skill so its first output is the list of what it could not determine.
Cannot determine from the notes:
- who signs off on acceptance, and in how many days
- whether the data migration is in scope or a
separate engagement
- which environment the work is delivered into
- what happens if the client's API is not ready
by week 3
Drafted the SOW with these marked TBD rather than
assumed. Four questions, one email.Four questions asked before the SOW goes out is the cheapest hour in consulting. The same four discovered in week five is a change order and a conversation nobody enjoys.
What goes into the skill's inputs
The notes, in whatever state they are.
Transcript, bullets, or a photo of a whiteboard typed up badly. Sorting is the skill's job.
Your clause library.
The exclusions, assumptions and acceptance language you have refined over years of engagements. The skill selects from this library. It does not compose new legal language, and it should be told so in as many words.
Two signed SOWs from work that went well.
Structure and level of detail come from these. Pick engagements that ended cleanly, not the ones that were most impressive.
Your definition of done.
How acceptance works at your firm: who reviews, how long they have, what happens on silence. Most SOW disputes are acceptance disputes wearing a different coat.
What it cannot do
- It cannot write the legal terms. Liability, indemnity, IP assignment and termination come from your lawyer and your clause library, unchanged.
- It cannot judge feasibility. A timeline that reads well and is impossible looks identical on the page to one that is merely tight.
- It cannot decide what to leave out. Scope is a commercial decision, and the exclusions list is where your margin lives.
- It cannot tell you the client will be difficult. You knew on the call; the notes will not say it.
Keep the SOW skill separate from the proposal skill
These two overlap enough that they will fight over the same requests, and whichever one wins will be the one whose description happens to contain the word you typed. Give each a boundary in its own description: the SOW skill handles scope, acceptance and exclusions, and hands pricing to the other. There is a separate guide on skills for writing client proposals, and one on writing a description that gets picked.
Works with Claude Code.
The Consultant pack
Thirteen skills for discovery notes, scoping, status reports and case studies, with the clause-library structure and the eval cases included. The pack page carries the price, the refund terms, and the date every skill was last tested.
See the packs