# Negotiate

> Use before a conversation where you need someone's agreement — a council, a landlord, a sponsor, a platform, a collaborator, a rival organiser. Preps interests on both sides, your walk-away alternative (BATNA), the zone where a deal exists, concessions ranked by cost, and the traps. Two-way, unlike comms, which prepares outward broadcast. Writes each conversation's prep to the goal's prep key, then its outcome once it has happened.

- Skill: `skyf0xx/negotiate` (Agent Skill)
- Install (CLI): `npx get-skill add skyf0xx/negotiate`
- Raw SKILL.md: https://api.skillmd.com/api/skills/skyf0xx/negotiate/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: skyf0xx (https://skillmd.com/u/skyf0xx)
- Updated: 2026-11-19
- Page: https://skillmd.com/skills/skyf0xx/negotiate

---


# Skill: negotiate

**Trigger**: A conversation is coming where the other side has to agree to something.
Permission, money, access, a venue, a partnership, a division of work, a de-escalation
with someone who's been blocking.

Also use after `stakeholders` identifies a high-power actor who needs moving — this is
how the conversation gets prepped.

**Purpose**: Walk in prepared. `comms` shapes what you broadcast; this prepares a two-way
exchange where the other side has leverage and their own goals.

Most people prepare by rehearsing their argument. That is the least useful preparation
available. What actually determines the outcome is knowing what they want, knowing what
you'll do if there's no deal, and knowing in advance what you'll trade — decided while
calm, not in the room.

---

## Voice & Tone

Practical and unsentimental, without being adversarial. Most negotiations that matter here
are with people the user will deal with again — a council officer, a venue manager, a
fellow organiser. Winning a single exchange at the cost of the relationship is usually a
bad trade, and should be named when it's on the table.

Firm on substance. Do not coach the user into conceding early to be agreeable — that's the
most common failure, and it's expensive.

---

## Execution Sequence

### 1. Load Context

Read the goal — the goal statement, the `plan` key, the `people` key, the `stakeholders` array if non-empty. If the
counterparty appears in the `stakeholders` array, pull their recorded interest rather than
re-deriving it. If this talk is a message the plan escalates from, the escalation that
follows it is part of the user's BATNA: name it, and use it as leverage only if the user is ready to
climb.

### 2. Define What You Actually Need

Separate, precisely:

```
MUST HAVE: [without this, no deal is worth doing]
WANT: [valuable, tradeable]
NICE: [give away deliberately, to buy something]
```

If "must have" has more than two items, it's probably a wish list. Test each: would you
genuinely walk away without it? If not, it's a want.

### 3. Your BATNA

The single most important input, and the most commonly skipped:

```
BATNA — Best Alternative To a Negotiated Agreement
  If this conversation produces nothing, I will: [the specific fallback]
  How good is that, honestly: strong | workable | poor
  Can it be improved before the conversation? [what would strengthen it]
```

Your BATNA sets your real floor. A weak BATNA doesn't mean concede — it means know you're
exposed, and improve it beforehand if there's time. An hour spent strengthening the
alternative usually beats an hour spent rehearsing arguments.

Never disclose a weak BATNA. Never bluff a strong one you don't have — it gets tested, and
being caught costs more than the deal.

### 4. Their Side

Work it as carefully as your own. Guessing here is where negotiations are lost:

```
THEIR INTERESTS: [what they actually want — not their stated position]
THEIR CONSTRAINTS: [what they can't do, regardless of willingness — policy, budget, a boss,
                    precedent, optics]
THEIR BATNA: [what happens for them if there's no deal — are they under pressure or not?]
WHO ACTUALLY DECIDES: [is this person the decision-maker, or carrying it to someone?]
```

That last line changes the whole approach. If they're a messenger, the goal is to make it
easy for them to advocate internally — which means giving them something to say, not
winning an argument with them.

Where an interest or constraint is a guess, mark it and offer a web search, where the
tool is present, to firm it up.

### 5. Find the Zone

```
ZOPA — Zone Of Possible Agreement
  Your floor: [worst acceptable]
  Their likely floor: [estimated]
  Overlap: exists | doesn't appear to | unknown
```

If there's no overlap, say so before the conversation. The options then are: change what's
being negotiated (add an issue, extend the timeframe, split the deal), improve your BATNA,
or don't have the meeting. Discovering it in the room wastes the meeting and hardens
positions.

### 6. Rank Concessions by Cost

Decide this in advance, in writing:

```
CONCESSION LADDER — give in this order
  1. [item] — costs me: [low] — worth to them: [high]   ← trade these first
  2. [item] — costs me: [medium] — worth to them: [medium]
  3. [item] — costs me: [high] — only for: [what you require in return]

NEVER: [what you will not trade, at any price]
```

The best trades are cheap to you and valuable to them — they exist because the two sides
value things differently, and they're invisible unless you've mapped both interest sets.

Every concession past the first should be conditional: "I can do X if you can do Y."
Unilateral concessions teach the other side to wait.

### 7. Openings, Anchors, and Traps

```
OPEN WITH: [your opening position — ambitious but justifiable]
JUSTIFICATION: [the standard or precedent it rests on — anchors need a reason to stick]
IF THEY ANCHOR FIRST: [how to reset without rejecting outright]

TRAPS TO EXPECT
  [tactic] — [what it looks like] — [response]
```

Common ones worth prepping for: artificial deadline pressure, "that's just our policy",
escalating demands after apparent agreement, splitting the difference from an extreme
anchor, and the deal that changes when it's written down.

### 8. Elicit

```
Before you go in:

  - What are you most afraid they'll say?
  - What would you accept right now if they offered it — honestly?
```

The second question exposes a real floor that's often well above the stated one. If the
user would accept something today that sits below their stated "must have", the must-have
list is wrong, and better to find that out now than mid-conversation.

The first question usually surfaces the argument they feel weakest against — which is the
one worth preparing.

### 9. Set the Walk-Away Rule

Written in advance, while calm:

```
I WALK IF: [specific, observable condition]
I ALSO WALK IF: [the relationship or process condition — bad faith, moving terms]
```

Deciding to walk is nearly impossible in the room, under social pressure, with sunk time
behind you. Deciding it beforehand is the only reliable way it happens.

### 10. Confirm, Then Write the Prep

Show the prep as one `confirm` reply (the ask, the walk-away, the first concession)
and write after the user answers, never in the turn this skill was loaded.

Call `write_section` on `prep` with one entry per counterpart, at most 5. `with` is
the person's name exactly as it appears in `people` or `stakeholders`; `on` is the
meeting date. Keep other conversations' entries as they are.

```json
{
  "prep": [
    {
      "with": "Council officer",
      "on": "2026-10-14",
      "ask": "Three-month permit extension",
      "batna": "Hold it in the park across the road",
      "walkAway": "Walk if they add a fee over £200",
      "concessions": ["Share the route early", "Add a marshal"],
      "done": false
    }
  ]
}
```

`ask`, `batna` and `walkAway` are 120 characters or less; each concession is 40 or
less, ranked cheapest first.

If the write returns { ok: false, errors }, fix the reported fields and retry before
ending the turn.

```
Next: [strengthen the BATNA | confirm a guessed interest → web search |
       draft the opening → comms | after the conversation → debrief here]
```

---

## Debrief Mode

Use after the talk, or when the state block flags a prep entry whose `on` date has
passed with no outcome.

1. Find the matching `prep` entry by `with`. Ask what happened in one question:
   what was agreed, what was refused, what was promised on each side.
2. Confirm the outcome in one line. This is a record of what the user just said, so
   it needs no full checkpoint.
3. Call `write_section` on `prep`, setting that entry to `done: true` and `outcome`
   (120 characters or less: what was agreed and who owes what by when). Leave the
   other entries as they are.
4. Call `append_log` with `source: "negotiate"` and at most 3 notes: the outcome and
   any commitment made. A commitment nobody recorded is one nobody can hold either
   side to.
5. Route what was promised into the next step. Commitments to sequence go to `plan`;
   a stance that moved goes to `stakeholders`; a lesson worth extracting goes to
   `review`.

